Authors: Shaina Raza, Mizanur Rahman, Safiullah Kamawal, Armin Toroghi, Ananya Raval, Farshad Navah, Amirmohammad Kazemeini
Page 1:
A C OMPREHENSIVE REVIEW OF RECOMMENDER SYSTEMS :
TRANSITIONING FROM THEORY TO PRACTICE
A P REPRINT
Shaina Raza∗ ∗1, Mizanur Rahman∗2, Safiullah Kamawal1, Armin Toroghi1, Ananya Raval1, Farshad Navah3, and
Amirmohammad Kazemeini1
1Vector Institute, Toronto, Canada , shaina.raza@vectorinstitute.ai,
safiullah.kamawal@vectorinstitute.ai, armin.toroghi@vectorinstitute.ai,
ananya.raval@vectorinstitute.ai, amirmohammad.kazemeini@vectorinstitute.ai
2Toronto, Canada , mizanur.york@gmail.com
3Toronto, Canada , farshad.navah@gmail.com
February 25, 2025
Contents
1 Introduction 2
2 Background 4
2.1 Historical Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Current State of Practice and Theory in Recommender Systems . . . . . . . . . . . . . . . . . . . . . 5
3 Literature Review Methodology 5
4 Challenges in Recommender Systems 7
5 Foundational Recommender Systems 8
5.1 Foundational Recommender Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
5.2 Can Foundational Recommender Systems address Practical Challenges? . . . . . . . . . . . . . . . . 10
6 The Era of Deep Learning in Recommender Systems 11
6.1 Deep Learning-based Recommender Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
6.2 Can Deep Learning-based Recommender Systems address Practical Challenges? . . . . . . . . . . . 12
7 Advanced Modeling Techniques in Recommender Systems 12
7.1 Graph-based Recommender Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
7.2 Sequential and Session-based Recommender Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 14
7.3 Knowledge-based Recommender Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
∗Equal Contribution.arXiv:2407.13699v2 [cs.IR] 23 Feb 2025
Page 2:
arXiv Template A P REPRINT
7.4 Reinforcement Learning-based Recommender Systems . . . . . . . . . . . . . . . . . . . . . . . . . 20
7.5 Large Language Model based Recommender Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 22
7.6 Multimodal Recommender Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
8 Specialized Recommender Systems 27
8.1 Context-aware Recommender Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
8.2 Review-based Recommender Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
8.3 Aspect-based Recommender Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
8.4 Explainable and Trustworthy Recommender Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 32
8.5 Fairness, Accountability, Transparency, and Ethics (FATE) in Recommender Systems . . . . . . . . . 33
8.6 Miscellaneous . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
9 Applications of Recommender Systems Across Different Domains 35
10 Discussion 37
10.1 Impact of this Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
10.2 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
10.3 Future Perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
11 Conclusion 39
ABSTRACT
Recommender Systems (RS) play an integral role in enhancing user experiences by providing
personalized item suggestions. This survey reviews the progress in RS inclusively from 2017 to 2024,
effectively connecting theoretical advances with practical applications. We explore the development
from traditional RS techniques like content-based and collaborative filtering to advanced methods
involving deep learning, graph-based models, reinforcement learning, and large language models. We
also discuss specialized systems such as context-aware, review-based, and fairness-aware RS. The
primary goal of this survey is to bridge theory with practice. It addresses challenges across various
sectors, including e-commerce, healthcare, and finance, emphasizing the need for scalable, real-time,
and trustworthy solutions. Through this survey, we promote stronger partnerships between academic
research and industry practices. The insights offered by this survey aim to guide industry professionals
in optimizing RS deployment and to inspire future research directions, especially in addressing
emerging technological and societal trends2. The survey resources are available in the public GitHub
repository https://github.com/VectorInstitute/Recommender-Systems-Survey .
Keywords Recommender Systems ·Graph-based Recommender Systems ·Knowledge-based Systems ·Multimodal
Recommender Systems ·Large Language Models ·Personalization ·Industry Applications ·Explainable AI ·
Transparency ·Fairness ·Deep Learning ·Survey
1 Introduction
Recommender Systems (RS) are a type of information filtering system designed to predict and suggest items or
content—such as products, movies, music, or articles—that a user might be interested in. These predictions are based
on the user’s past behavior, preferences, or the behavior of similar users [ 1]. The main goal of any RS is to enhance
user experience, increase engagement, and facilitate decision-making processes [ 2]. This is applicable across various
domains, including e-commerce, entertainment, and social media. RS hold significant roles in both theoretical research
(academics) and practical applications (industry).
2The ideas and discussions produced in this work are strictly those of the authors and do not represent the points of view of the
institutions the authors belong to.
2
Page 3:
arXiv Template A P REPRINT
The importance of RS has grown exponentially with the advent of big data and advancements in artificial intelligence
[3,4]. As users interact with digital platforms, they generate vast amounts of data that can be leveraged to make precise
and personalized recommendations. This ability to tailor suggestions not only improves user satisfaction but also
increases the likelihood of users discovering new and relevant content [ 5]. In e-commerce, for example, RS can drive
significant sales by suggesting products that align with users’ preferences, while in entertainment, they enhance user
engagement by recommending shows or music that match users’ tastes. Additionally, RS are now being integrated into
new and emerging fields such as personalized education [ 6], where they help tailor learning experiences to individual
student needs, and healthcare [ 7], where they assist in suggesting personalized treatment plans and health interventions.
The development of large language models (LLMs) further enhances RS by enabling them to understand and process
vast amounts of text data, leading to more sophisticated and context-aware recommendations [8].
In academia, RS are the subject of extensive research aimed at understanding user behavior and decision-making
processes [ 3]. This research utilizes sophisticated data analytics and Artificial intelligence (AI) techniques. The ACM
Recommender Systems Conference (RecSys) [ 9], along with related scholarly journals and venues, highlights emerging
technologies and their potential impact across various sectors, including entertainment, e-learning, and academic
publishing. Recent academic advancements have focused on integrating reinforcement learning and LLMs into RS,
leading to more accurate and dynamic recommendation capabilities.
In the industry, RS enhance customer satisfaction and drive revenue growth by providing tailored suggestions [ 5]. Major
corporations such as Amazon, Netflix, and Spotify integrate RS into their operations, significantly contributing to their
business models. For instance, Amazon reports that 35% of its revenue comes from its RS [ 10], while Netflix attributes
revenues of approximately $33.7 billion and its success in customer retention significantly to its RS [ 11]. The global
market for recommendation engines, as per Precision Reports [ 12], is forecasted to witness substantial growth from
2023 to 2030, highlighting their increasing importance in business strategies. Privacy-preserving algorithms and bias
mitigation are also becoming key areas of focus for industry practitioners.
This survey focuses on the theory of RS and their transition to practical applications, aiming to bridge the gap between
academic research and industry practices. It highlights how theoretical advancements can be effectively implemented in
real-world scenarios.
Necessity of this Survey
Previous surveys often concentrate solely on the theoretical aspects of RS, exploring methods and algorithmic founda-
tions to improve prediction accuracy and personalization [ 13,4]. Conversely, practical-focused research or applications
typically views RS as essential tools for enhancing user engagement, retention, and business growth [ 14,15]. There is a
need for the collaboration between academia and industry to address both technical challenges and real-world demands,
which in turn enhances user satisfaction and business value. This interdependence highlights the growing importance of
such partnerships.
Difference with Existing Surveys
Unlike previous surveys that often focus solely on the theoretical or practical aspects, our survey uniquely covers the
integration of theoretical advancements with practical applications, offering a comprehensive overview that addresses
both academic and industry perspectives. Furthermore, we identify emerging trends and future research directions, such
as the integration of explainable AI in RS to ensure transparency and user trust.
Table 1: Overview of Related Surveys Ordered by Date of Publication and Comparison Criteria
Survey Topic Theory Practise Survey Topic Theory Practise
[16] Economics X ✓ [17] Stock market X ✓
[18] Digital marketing ✓ ✓ [19] Finance ✓ X
[20] Multimedia content ✓ X [21] Travel X ✓
[22] Health ✓ ✓ [7] Health ✓ X
[23] Health ✓ X [24] Health ✓ X
[25] Health X ✓ [26] Health ✓ X
[27] E-learning ✓ X [6] E-learning ✓ ✓
[28] E-Learning X ✓ [29] Machine learning ✓ X
[30] Knowledge integration ✓ X [31] Explainability ✓ X
[32] Context awareness ✓ X [33] Context awareness ✓ X
[34] Collaborative filtering ✓ X [35] Collaborative filtering ✓ X
[36] Hybrid methods ✓ X [37] Sequence awareness ✓ X
Continued on next page
3
Page 4:
arXiv Template A P REPRINT
Table 1: Continuation of Survey List
Survey Topic Theory Practise Survey Topic Theory Practise
[38] Session integration ✓ X [39] Session integration ✓ X
[40] Conversation integration ✓ X [41] Music ✓ ✓
[42] Music ✓ X [43] Reinforcement learning ✓ X
[44] Adversarial methods ✓ X [45] Review texts ✓ X
[46] Graph neural network ✓ X [47] Graph Neural network ✓ X
[48] Graph Neural network ✓ X [49] Deep learning ✓ X
[8] Large Language Models ✓ X [50] Large Language Models ✓ X
[51] Large Language Models ✓ X [52] Large Language Models ✓ X
[53] Large Language Models ✓ X [54] Large Language Models ✓ X
[55] Large Language Models ✓ X [56] Aspect integration ✓ X
[57] General ✓ X [4] General ✓ X
[58] News X ✓ [59] News ✓ X
[60] News ✓ X [61] Privacy ✓ X
[62] Tourism ✓ X [63] Evaluation ✓ X
[3] General ✓ X [64] General ✓ X
[65] Trustworthiness ✓ X [66] Cultural Heritage X ✓
Difference: Our survey covers the theory of RS and the application of its methods in practice.
Main Contributions
1.This survey provides a comprehensive review of RS, tracing their development from theoretical foundations to
practical applications between 2017 and 2023. It is the first survey to specifically highlight the translation of
theoretical advancements into practical solutions for industry challenges.
2.Each type of RS is thoroughly examined, including data input methods, associated challenges, relevant datasets,
evaluation metrics, model accuracy, and practical applications, as presented in tables. The survey aims to offer
industry professionals a set of guidelines to facilitate the deployment of these systems in real-world settings.
3.We discuss the specific challenges faced by RS in various sectors, such as e-commerce, healthcare, finance, and
others. The survey emphasizes the need for scalable, real-time, and privacy-focused solutions, demonstrating
how theoretical insights can address these industry-specific demands.
2 Background
Recommender systems (RS) are algorithms designed to suggest items—such as books, movies, products, or content—to
users based on their preferences. The primary goal of RS is to enhance user experience by personalizing content [ 3]. At
its core, an RS combines user and item profiles with a filtering mechanism to align user preferences with suitable items
[13]. User profiles gather data such as demographics and browsing history, while item profiles detail features like genres.
Both explicit feedback (e.g. ratings) and implicit feedback (e.g. browsing actions) refine these recommendations.
2.1 Historical Context
One of the pioneering efforts for RS is from Elen Rich in 1979 [ 67] to suggest books based on user preferences
categorized into “stereotypes". Following this, Jussi Karlgren conceptualized the “digital bookshelf" in 1990 [ 68], an
idea later expanded by researchers at SICS, MIT, and Bellcore, with notable contributions from Pattie Maes, Will Hill,
and Paul Resnick, whose GroupLens project [ 69] received the 2010 ACM Software Systems Award. Later, Adomavicius
[3], Herlocker [70], and Beel [71] provided foundational theory on RS.
Traditional RS methods can be categorized into collaborative filtering, content-based filtering, and hybrid approaches,
aiming to improve user experience [ 13]. Collaborative filtering (CF) [ 70] is based on the idea that users with similar
preferences will likely have similar tastes in the future. CF recommends items by finding a neighborhood of similar
users or items. CF can recommend items without needing much content analysis, however, it normally faces challenges
like cold starts, scalability, and sparsity [ 34] . Content-based filtering (CBF) [ 72] recommends items based on a user past
preferences and item characteristics, using techniques like Term Frequency - Inverse Document Frequency (TF-IDF),
cosine similarity, and neural networks for item representation. However, it may struggle with recommending new or
unseen items. Hybrid RS [ 36] combine the strengths of both approaches, offering more accurate and personalized
recommendations by integrating diverse methodologies.
4
Page 5:
arXiv Template A P REPRINT
The Netflix Prize [ 73], a competition aimed at enhancing RS algorithms, significantly popularized these algorithms.
While the competition focused on improving accuracy, an essential aspect of algorithmic effectiveness, it also empha-
sized the importance of diversity, privacy, and serendipity in boosting user satisfaction [63].
Machine Learning (ML) methods such as k-nearest neighbors algorithm (k-NN), deep neural networks, and Natural
Language Processing (NLP) have enhanced RS over the years by providing more precise recommendations. However,
key challenges and ethical issues, such as safeguarding user and data privacy, mitigating biases for fair recommendations,
transparency for user trust, and keeping pace with technological advancements remain the challanges.
2.2 Current State of Practice and Theory in Recommender Systems
Academia focuses on the theory, methods, and algorithms in RS, while the industry emphasizes practical applications,
scalability, and direct business impacts. This section explores the distinct challenges faced by these two sectors.
Theoretical Research on Recommender Systems Theoretical research on RS is commonly initiated by academics
through the development of new algorithms, models, and evaluation metrics. However, academic researchers face
challenges in accessing diverse and comprehensive datasets due to privacy concerns, proprietary restrictions, and
financial barriers. Additionally, data quality issues such as biases, inaccuracies, and outdated information limit the
development and testing of RS in varied contexts.
The drive for high accuracy in research models can often lead to overfitting, which makes them unusable for real-world
applications. Such a focus may neglect crucial aspects like diversity, novelty, and user satisfaction. Additionally,
solutions from academia are frequently not easily adaptable in industry settings due to their reliance on data-intensive
algorithms, complexity, and a disconnect in keeping developers updated.
Practices in Recommender Systems The industry faces several challenges in deploying RS, particularly concerning
scalability as user bases and catalog sizes expand. Adapting to constantly evolving user preferences and content
availability presents ongoing difficulties. Ensuring the diversity and fairness of recommendations is crucial to avoid
biases. Additionally, integrating real-time data and maintaining high performance under heavy loads are significant
challenges. Balancing personalization with privacy concerns requires careful handling of user data to build trust and
comply with regulations.
Common Challanges Both theory and practice emphasize the importance of high-quality (accurate, relevant, reliable,
and representative of the intended use case or application) datasets for building RS. Academic research often relies
on high-quality data for benchmarking purposes, while the industry frequently requires such data to enhance user
experience and system effectiveness.
Theoretically, RS algorithms are quite advanced now, featuring layers of deep neural networks and the latest language
model complexities. In practice, however, these models are not immediately applicable to real-world use cases. Industry
sectors, generally running a set of standard models, require significant adaptations to implement these advanced
algorithms effectively
In this survey, we examine the theoretical and practical aspects of RS, with the goal to facilitate a smooth transition
from research to real-world application.
3 Literature Review Methodology
To compile a comprehensive and relevant list of papers for our review, we conducted a systematic literature review,
adhering to established methodology principles [74]. Our search query and extraction methods are detailed below.
Research Questions
1. How have RS algorithms evolved theoretically over the years?
2. What strategies can be utilized to apply theoretical advancements in RS to practical applications?
Databases Searched For selecting studies, we gathered articles published inclusively from January 2017 through
April 2024. This timeframe was chosen because much of the evolution in RS is linked to deep learning algorithms,
also highlighted in the RecSys workshop in 2017 [ 75]. We conducted our literature search across multiple academic
databases and digital libraries renowned for their extensive collections of RS literature, including IEEE Xplore, ACM
Digital Library, PubMed, ScienceDirect, JMLR, and Wiley. To refine the search results, we applied specific inclusion and
5
Page 6:
arXiv Template A P REPRINT
exclusion criteria based on the publication year, relevance to RS, the source, and the paper’s focus on the technological,
theoretical, and application aspects of RS. Only peer-reviewed journal articles, conference papers, and significant arXiv
papers were considered.
Search Query Our search strategy aimed to find literature across various aspects of RS, including types, algorithms,
evaluation, application, user interaction, and data quality. The search query used was:
("recommender systems" OR "recommendation systems" OR "RS" OR "RecSys") AND ("content-based filtering"
OR "collaborative filtering" OR "hybrid recommender systems" OR "context-aware recommender systems" OR
"knowledge-based systems" OR "social recommender systems") AND ("matrix factorization" OR "deep learning" OR
"convolutional neural networks" OR "recurrent neural networks" OR "reinforcement learning" OR "autoencoders" OR
"neural collaborative filtering" OR "graph neural networks") AND ("precision" OR "recall" OR "F1 score" OR "RMSE"
OR "MAE" OR "hit rate" OR "novelty" OR "diversity" OR "serendipity" OR "user satisfaction") AND ("e-commerce"
OR "media streaming" OR "social media" OR "education" OR "healthcare" OR "tourism" OR "personalized news" OR
"job recommenders") AND ("user interface" OR "user experience" OR "usability" OR "interaction design" OR "user
engagement" OR "user feedback" OR "user profiling") AND ("explicit feedback" OR "implicit feedback" OR "data
sparsity" OR "cold start problem" OR "data quality" OR "user-generated content") AND ("privacy" OR "data security"
OR "ethical algorithms" OR "bias and fairness" OR "transparency" OR "recommendation explainability") AND ("tech
industry" OR "startup case studies" OR "market analysis" OR "business models" OR "return on investment" OR "user
retention") AND ("transformer models" OR "BERT" OR "GPT" OR "natural language understanding" OR "language
generation" OR "sentiment analysis" OR "text embeddings") AND ("user personalization" OR "adaptive systems" OR
"customization techniques" OR "user-adaptive content" OR "dynamic personalization")
Table 2: Inclusion and Exclusion Criteria for Literature Review
Inclusion Criteria Exclusion Criteria
Articles that include at least 3-4 keywords from our search
query in the title, abstract, or keywords.Articles without relevant keywords.
Articles published from 2017 through January 2024. Articles published outside this timeframe, except classical
papers that need to be cited.
Articles that pass the initial screening based on titles and
abstracts and address RQ1 or RQ2.Articles that do not address RQ1 or RQ2.
Articles from published work or arXiv if it covers an impor-
tant and relevant topic.Grey literature, e.g., technical reports, or dissertations.
Articles in English language. Articles not written in English.
Quality Assessment We screened the articles using the inclusion and exclusion criteria detailed above in Table 2. If
there was any uncertainty, the paper was briefly reviewed and then either included or excluded based on consensus
from the two first co-authors. Selected papers underwent a thorough reading and were subject to a quality assessment
involving a set of questions:
A paper was considered for inclusion in our review if it received a “yes" or “partially" response to any of these questions:
1. Does the article present a novel RS or methodology?
2. Is the article related to a RS in an academic setting?
3. Is the article related to a RS in an industry setting?
4. Does the article propose a framework, tool or methodology?
5.Has the research provided a concise statement or definition outlining its aims, goals, purposes, problems,
motivations, objectives, and questions?
The results of the literature search were finalized and categorized by database, as shown inTable 3.
Bibliometric analysis Figure 1 presents a comprehensive analysis of publications reviewed in this survey. The top
left chart shows the distribution of articles by type, highlighting the dominance of experimental studies. The top right
chart details the publication trends by year. The bottom left chart lists the top journals, with ’ACM Transactions on
Information Systems’ leading in terms of publication count. The bottom middle chart displays the top conferences, with
the ’ACM SIGIR Conference on Research and Development in Information Retrieval’ being the most frequented.
6
Page 7:
arXiv Template A P REPRINT
Table 3: Identified Papers by Database
Publisher/Journal/Conference Number of Papers
ACM 83
IEEE 43
Springer 32
ScienceDirect 25
arXiv 15
Others 89
Figure 1: Overview of Publication Trends and Key Venues in RS Research.
4 Challenges in Recommender Systems
RS play a vital role in personalizing user experiences and driving business value across various domains. Despite their
widespread adoption, several challenges persist in their deployment and maintenance.
E-commerce E-commerce platforms face the challenge of personalizing the shopping experience by recommending
products in real-time, managing vast data, and adapting to changing consumer preferences [ 76]. Personalization must
consider factors such as time, season, location, and the user’s current situation. For example, recommendations for a
new parent shopping for baby products will differ significantly from those for a book lover. Introducing diversity and
novelty in recommendations is crucial to keep the experience fresh and engaging.
7
Page 8:
arXiv Template A P REPRINT
Entertainment In the entertainment industry, the challenge lies in tracking users’ preferences across genres while
introducing them to new content to maintain engagement. Balancing personalization and novelty is essential. Music
recommendations require frequent updates due to the shorter shelf life of songs compared to movies [ 77]. Conversely,
movie recommendations are less dependent on frequent updates, as films typically have a longer shelf life. However,
effective movie RS should still balance between promoting new releases and maintaining a selection of enduring
favorites to satisfy a wide range of user preferences [78].
News The news industry must deliver personalized content promptly without overwhelming users. News preferences
are highly dynamic, necessitating recommendations that adapt to rapid changes in interests and current events [ 59]. It is
important to offer diverse viewpoints to prevent echo chambers, combat misinformation [79], and maintain user trust.
Tourism Personalized booking recommendations in tourism must account for user preferences regarding destinations,
travel dates, budgets, and accommodations [ 80]. Integrating factors such as past travel history, seasonal trends, and
real-time availability is essential. Balancing immediate needs, like dining recommendations during travel, with advance
bookings for stays and major attractions enhances the overall user experience.
Healthcare Healthcare RS face issues like data privacy, security, and patient consent under regulations such as Health
Insurance Portability and Accountability Act (HIPAA). Providers, patients, and administrators all require access to
relevant information tailored to their roles, necessitating role-based solutions and robust processing capabilities to
handle large volumes of heterogeneous data effectively [81].
Finance Financial RS need to navigate data privacy, security, and compliance with regulations like General Data
Protection Regulation (GDPR) and Payment Card Industry Data Security Standard (PCI-DSS). Challenges include
managing data quality, integrating diverse data sources, and providing personalized financial advice [ 19]. Ensuring the
interpretability and transparency of recommendations is crucial for building trust and user confidence. Fairness of the
recommendations is of utmost importance.
E-learning E-learning RS face the task of addressing varied user needs, overcoming the cold start problem with
new users, and handling data sparsity. Adapting to dynamic content and user behavior, ensuring contextual relevance,
scalability, and employing suitable evaluation metrics to assess educational impact are fundamental challenges [82].
Discussion RS across various sectors face a set of common challenges despite their industry-specific characteristics.
Generally, these systems struggle with balancing personalization and user privacy, managing data scalability, and
ensuring the diversity and novelty of recommendations to keep users engaged. They must also address the cold-start
problem, where insufficient user data can hinder the system’s ability to make accurate recommendations. Additionally,
dynamic user preferences require systems to continually adapt and learn from new data, posing challenges in real-time
processing and algorithmic efficiency. Lastly, ensuring fairness and avoiding bias in recommendations is crucial, as
these systems often influence user decisions and can perpetuate existing disparities if not carefully managed.
In the following sections, we explore the evolution of RS and how they address these challenges.
5 Foundational Recommender Systems
A RS can be mathematically represented as a function fthat predicts the utility of an item ifor a user u, denoted as ˆrui,
which estimates how much user uwould prefer item i.[3] . This function is typically learned from historical data:
ˆrui=f(u, i; Θ) (1)
where Θrepresents the parameters of the model, learned from the data. In the context of RS, the term ˆruirepresents a
prediction of the rating or utility that a user would assign to an item This prediction is used for recommending items
that are likely to be of interest to the user. A general framework of RS is illustrated in Figure 2. The lifecycle of a
data-driven model within an RS starts with data acquisition, followed by storage and preparation. This leads to feature
engineering, forming the basis of the data pipeline. The data pipeline feeds into the training pipeline, which includes
model training and validation. Following training, the process involves candidate generation and ranking. This process
is complemented by A/B testing, offline and/or online evaluation. The final stages include deployment and monitoring.
Below, we present an overview of foundational RS.
8
Page 9:
arXiv Template A P REPRINT
Figure 2: General Framework of RS showing an outline of a data processing pipeline. It covers stages from data
acquisition, preparation, and feature engineering, through model training and validation, to deployment and monitoring.
The pipeline also includes steps for candidate generation, ranking, evaluation, and a feedback loop to ensure continuous
improvement and business value.
5.1 Foundational Recommender Systems
Foundational RS refers to the early models and techniques that established the core principles and methodologies in the
field of recommendation engines. These systems primarily include collaborative filtering, content-based filtering, and
hybrid approaches.
Content-Based Filtering Content-based filtering (CBF) is a recommendation strategy that suggests items to a user
based on the attributes of the items and a profile of the user’s preferences, typically utilizing similarity measures to
match user preferences with item attributes [ 83]. In CBF, the recommendation ˆruiis based on the features of items ϕ(i)
and a profile of the user’s preferences θ(u):
ˆrui=θ(u)⊤ϕ(i) (2)
•ϕ(i)represents the feature vector of the item.
•θ(u)represents the user preference vector.
The evolution of CBF recommendation models starts from traditional methods [ 83] like vector-space models, proba-
bilistic models, and decision trees, relying on manual feature engineering and similarity calculations. Vector-space
models compute item similarity through cosine similarity [ 84], probabilistic models estimate the likelihood of user
preference with statistical analysis [ 85], and decision trees recommend items by categorizing them based on attributes
[83]. These models laid the groundwork for personalized recommendations by leveraging explicit item features and
user preferences.
Algorithmic advancements in computer science and ML methods led to a shift to sophisticated neural networks that
allow automatic feature extraction and learning complex data patterns [ 86]. These neural network-based systems
leverage deep learning to analyze user interactions and item features across different modalities, including textual (such
as reviews [ 87], citations [ 88], and news [ 89]), streaming (like music [ 77]), and image data [ 90]. The overall goal of
these advancements is to generate personalized recommendations by aligning user profiles with item characteristics.
Challenges with CBF: In general, CBF faces challenges such as the cold-start problem, over-specialization (only
suggesting items similar to those the user has already seen or liked), computational cost (which can increase quadratically
or cubically with the number of users and items), and a lack of updates to user and item profiles.
Collaborative Filtering Collaborative filtering (CF) is a technique used by RS to predict the preferences of a user
based on the preferences of similar other users [ 34]. CF techniques are broadly divided into two main categories:
memory-based and model-based methods. Memory-based CF makes recommendations using similarities between users
or items directly from user ratings. It has further two types: user-based CF predicts a user ratings based on similar
users’ historical ratings [ 70], while item-based CF predicts ratings based on similar items [ 91]. Both methods face
9
Page 10:
arXiv Template A P REPRINT
Figure 3: Timeline of Foundational Recommender Systems.
challenges such as scalability and sparsity. Model-based CF, like matrix factorization [92] and factorization machines
[93], uncover latent factors representing user preferences and item characteristics. These methods decompose the
user-item interaction matrix into latent feature vectors for users and items.
CF often starts with constructing a user-item interaction matrix Rwith users, items, and ruirepresenting known
interactions between users and items. One popular approach within CF is matrix factorization [ 92], where Ris
approximated by the product of two lower-dimensional matrices U(user features) and I(item features):
ˆR=U⊤I (3)
•Uis ak×mmatrix, with kbeing the number of latent factors and mthe number of users.
•Iis ak×nmatrix, with nbeing the number of items.
Neural extensions in CF have advanced these RS, utilizing deep learning to capture intricate user-item relationships
and significantly improve recommendation accuracy. Techniques such as Neural Collaborative Filtering (NCF) [ 94],
Sequence-Aware RS [ 95], and Graph Neural Networks (GNNs) [ 48] have emerged as state-of-the-art approaches in CF.
Challanges with CF Like CBF, these methods are more accurate and robust, however they also present challenges such
as computational complexity and limited interpretability, which may hinder their scalability and practical applicability
in real-world scenarios.
Hybrid Approaches A hybrid RS combines multiple recommendation techniques, such as CBF, CF and other ML
models to improve the accuracy and relevance of recommendations provided to users. The most common hybrid
techniques include weighted combination, switched selection, feature combination, cascading, and feature augmentation
[96]. The combination can be represented as a weighted sum:
ˆrui=α·fCB(u, i; ΘCB) +β·fCF(u, i; ΘCF) (4)
•fCBandfCFrepresent the content-based and collaborative filtering functions, respectively.
•ΘCBandΘCFare the parameters for each respective model.
•αandβare weights that balance the contribution of each method.
Techniques such as the Wide & Deep Learning framework [ 97], Neural Factorization Machines (NFM) [ 98], DeepFM
[99], and Deep & Cross network [ 100] combine explicit feature interactions and implicit feature hierarchies, leveraging
both shallow and deep learning models for enhanced recommendations.
A timeline illustrating the evolution in foundational RS is given in Figure 3. Prominent publications under foundational
RS are given in Table 4.
5.2 Can Foundational Recommender Systems address Practical Challenges?
Foundational RS, which include CF, CBF, and hybrid methods, form the core of many personalized recommendation
solutions. While these systems have proven effective across various industries, their ability to address practical
10
Page 11:
arXiv Template A P REPRINT
Table 4: Publications on Foundational RS.
Method Publications
CF [69, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112, 100,
113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 98, 125,
126, 127, 128, 43, 129, 130, 131, 132]
CBF [133, 134, 135, 136, 101, 32, 137, 138, 127, 83, 90, 139, 132]
Hybrid [36, 140]
challenges in sectors like e-commerce, entertainment, news, tourism, finance, healthcare, and e-learning is often limited.
For example, CF can personalize user experiences by leveraging behavior data in a news RS [ 59] or a music RS [ 42],
but struggles with the dynamic nature of user preferences and the need for real-time recommendations. In tourism,
finance, healthcare, and e-learning, foundational RS can understand user preferences and behavior patterns, but issues
like the cold start problem, data diversity, privacy concerns, and the need for highly personalized services require more
sophisticated solutions. These solutions often blend foundational techniques with modern advancements like deep
learning and specialized RS (discussed below).
6 The Era of Deep Learning in Recommender Systems
In recent years, deep learning has emerged as the standard in RS, as detailed in a related survey [ 86]. In the context of
this discussion, we shed some light on some popular deep neural network based RS.
6.1 Deep Learning-based Recommender Systems
Multi-Layer Perceptrons Traditional RS primarily use linear methods like matrix factorization [ 92], which struggle
with capturing complex user-item interactions. In contrast, Multi-Layer Perceptrons (MLPs), a type of feedforward
neural network, use deep layers to model these nonlinear interactions more accurately, improving both prediction
accuracy and recommendation quality. The evolution of MLPs is seen in RS such as Neural Collaborative Filtering (NCF)
[94], Deep Factorization Machine (DFM) [ 99], Wide & Deep [ 97], xDeepFM [ 141], Deep & Cross Network (DCN)
[100], FMLP-Rec [ 142], a model with learnable filters for improving sequential recommendation—and FinalMLP [ 143],
which combines dual MLP architectures with feature selection for effective Click-Through Rate (CTR) prediction.
Challenge with MLP: Despite their success, MLP models in RS face challenges like complexity, the risk of overfitting,
lack of spatial invariance, issues with vanishing or exploding gradients, and explainability concerns.
Autoencoders are neural network architectures specifically designed for unsupervised learning and are used for an
effective dimensionality reduction method. An autoencoder comprises two components: an encoder for compressing
input data into a lower-dimensional representation, and a decoder, which reconstructs the original data. Unlike traditional
MLP models, autoencoders explicitly capture this encoding-decoding structure.
Notable RS include AutoRec [ 144], Collaborative Filtering Autoencoder, Multi-Variational Auto-Encoder (Multi-
V AE) [ 145], Deep Recommender (DeepRec) [ 146], Recommender Variational Auto-Encoder (RecV AE) [ 147], Item-
based variational auto-encoder for fair recommendation [ 148], and the Variational Bandwidth Auto-Encoder (VBAE)
hybrid RS [ 149] . These approaches address sparsity and noise challenges, making them effective for personalized
recommendations.
Challenge with Autoencoder: Autoencoders are powerful for dimensionality reduction and capturing complex data
structures, but one key issue is their sensitivity to noise, which can lead to poor reconstructions if the input data is noisy
[150]. Also, the reconstruction process might not always preserve meaningful patterns essential for recommendations.
Convolutional Neural Networks (CNNs) can learn from visual, sequential, and multimodal data and have enhanced
accuracy and personalization of recommendations. CNNs have been applied in RS in various settings. DeepCoNN
analyzes text and visual cues to understand user preferences [ 87,151]. CNNs are integrated with graph structures
for scalable recommendation systems [ 120], employed in DKN for news recommendations [ 152], and utilized in
MusicCNN for music recommendations based on audio signals [ 153]. CNN-based RS models predict next-item
recommendations [ 130], recognize user preference patterns through CoCNN [ 154], and leverage collaborative filtering
with CAGCN [155].
Challenge with CNNs : CNNs in RS face challenges such as data sparsity, scalability, privacy, and domain-specific
issues [156]. Researchers continue to explore solutions to enhance CNN-based RS performance and usability
11
Page 12:
arXiv Template A P REPRINT
Table 5: Deep Learning-based RS Publications
RS Type Publications
GNN [48, 134, 169, 170, 171, 64, 172, 47, 173, 30, 174, 175, 176, 177, 178,
179, 118, 180, 181, 182, 183, 184, 185, 186, 187, 188, 189, 120, 190,
54, 191, 46, 192, 193, 122, 194, 195, 196, 197, 198, 199, 200, 201,
202, 203, 204, 205, 206, 207, 208, 209, 210, 211, 157, 212, 213, 214,
215, 216, 217, 168, 155, 89, 218, 219, 194, 220, 221, 222, 223, 224]
Sequential [225,38,226,227,228,163,179,39,117,181,186,187,229,230,160,
165,159,231,232,161,233,37,234,95,235,236,216,158,237,238,
239, 142, 240, 219, 241, 242, 243, 244, 245, 246, 167, 247, 248, 249]
KG [173, 30, 176, 180, 183, 196, 197, 198, 199, 206, 157, 212, 213, 215,
168, 89, 218, 221]
RL [250, 251, 106, 252, 253, 254, 255, 256, 257, 258, 259, 260, 261, 119,
230, 262, 263, 264, 265, 266, 267, 268, 269, 270, 214, 271, 43, 272,
273, 274, 275, 276, 277, 217, 278, 279, 167, 280, 281, 223]
LLM [282, 51, 283, 284, 285, 286, 287, 288, 289, 50, 290, 291, 292, 293, 8,
294, 295]
Multi-modals [296, 204, 297, 298, 299, 300, 249, 301, 218, 243, 244, 302]
Recurrent Neural Networks (RNNs) are adept at capturing complex user-item interactions within sequential data
[157]. The evolution of RNN-based RS began with GRU4Rec, utilizing Gated Recurrent Units (GRUs) for session-
based recommendations [ 158,159]. NARM introduced an attention mechanism to enhance accuracy [ 160], while
SASRec used self-attention to capture long-term semantics [ 161]. Deep attention neural networks were employed for
session-based recommendations [ 162]. RNNs, including LSTMs and GRUs, comprehend temporal dynamics in user
behavior [ 163,164]. Integrating RNNs with CNNs via recurrent convolutional networks offers deeper insights into
user preferences [ 165], followed by CNN-RNN hybrid RS [ 166] and Reinforcement Learning-based cross-domain
recommendations [ 167]. Additionally, a knowledge graph recommendation algorithm using RNN encoders has emerged
[168].
Chalenge with RNNs : A common challenge with RNNs is exploding and vanishing gradients [ 49]. Addressing these
issues often requires careful initialization, gradient clipping, or alternative architectures like LSTM networks that can
mitigate gradient problems. Additionally, training is sequential, as RNNs takes the data in a sequential manner, unlike
CNNs.
Table 5 shows the main publication based on deep learning RS.
6.2 Can Deep Learning-based Recommender Systems address Practical Challenges?
Deep learning-based RS have effectively addressed many practical challenges faced by foundational systems. Models
like NCF, DeepFM, and DeepMF have enhanced personalization by capturing complex user-item interactions in
e-commerce product recommendations [ 303]. Wide & Deep Learning has shown improved performance in e-commerce
for both product recommendations and ads [ 304]. CNNs, RNNs, and their variations and hybridizations are used in
content-based and sequential data recommendations, benefiting industries like news [ 164] and entertainment [ 305].
Transformer models like BERT are used for movie recommendations [ 228]. GNNs capture relationships in social
networks and e-commerce, offering improved accuracy and diversity in recommendations [155].
Advancements in deep learning have brought further changes to the theory and practice of RS, leading to the development
of advanced modeling methods, which is discussed next.
7 Advanced Modeling Techniques in Recommender Systems
7.1 Graph-based Recommender Systems
Graph Neural Networks (GNNs) are specialized neural networks designed to work with graph-structured data. GNNs
have emerged as a powerful tool in RS due to their capability to efficiently leverage complex, relational user-item
interaction data, enhancing recommendation accuracy and personalization. GNNs in RS are highlighted in the related
survey articles [48, 47], showcasing their significant impact and evolution in the domain.
12
Page 13:
arXiv Template A P REPRINT
In RS, a graph G= (V, E)represents the domain, with Vdenoting nodes (users and items) and Erepresenting
user-item interactions. Each node v∈Vis associated with a feature vector xv. GNN-based models adapt to various
graph types, including homogeneous (edges link nodes of a single type), heterogeneous (nodes and edges of multiple
types), and hypergraphs (edges connect more than two nodes). The core operation in GNNs, message passing, involves
nodes aggregating and updating information from neighbors to refine their features, thus capturing the dynamics of
user-item interactions. This process enhances recommendation accuracy and personalization by utilizing the relational
data within RS. The update mechanism for a node vat layer lis given by [47]:
h(l+1)
v =UPDATE(l)
h(l)
v,AGGREGATE(l)n
h(l)
u:u∈ N(v)o
(5)
Here, h(l)
vrepresents node v’s feature vector at layer l,N(v)denotes v’s neighbors, and UPDATE(l)and
AGGREGATE(l)are the respective update and aggregation functions. The objective of GNN-based RS is to learn a
predictive function ffor estimating the interaction likelihood between user uand item i, utilizing their feature vectors
huandhi:
ˆyui=f(hu,hi; Θ) (6)
In this context, ˆyuiis the interaction prediction score, with Θindicating the model parameters. Training involves
minimizing a loss function Lthat compares predicted scores ˆyuiwith actual interactions yui:
L=X
(u,i)∈Dloss(ˆyui, yui) (7)
This equation reflects the sum of losses over all observed user-item interactions in set D.
State-of-the-art RS models using GNNs GNNs have progressively transformed RS, starting from the foundational model,
i.e. Graph Convolutional Matrix Completion (GCMC) [ 189], which applies deep learning to user-item interaction
graphs for effective link prediction. Building upon this, GraphSAGE [ 194] is an inductive framework utilizing node
features for dynamic environments, though it could not address the complexity of real-world interaction data. Pinterest’s
PinSage [ 120] is a scalable model for web-scale graphs, improving its predecessor model for handling billions of nodes.
The Neural Graph Collaborative Filtering (NGCF) model [ 208] combines collaborative signals into user and item
embeddings, enhancing recommendation quality at the expense of increased complexity. Knowledge Graph Attention
Network (KGAT) [ 197] integrates knowledge graphs, improving recommendation diversity and explainability. The
Heterogeneous Graph Attention Network (HGAT) [ 122] incorporates hierarchical attention into the RS, addressing the
heterogeneity in relationships and node types.
Feature Interaction Graph Neural Networks (Fi-GNN) [ 184] represented a shift towards capturing multifield feature
interactions, notably in CTR prediction. Concurrently, Session-based Recommendation Graph Neural Network
(SR-GNN) [ 216] tackled session-based recommendations, enhancing accuracy by capturing item transitions. The
Multi-Modal Graph Convolution Network (MMGCN) [ 204] integrates multi-modal data into the graph-based learning,
though facing scalability challenges.
The introduction of LightGCN [ 202], with its focus on neighborhood aggregation and streamlined architecture, repre-
sents a simplification in the GNN landscape, improving efficiency without compromising performance. MixGCF [ 203]
brought forward a novel approach to negative sampling, optimizing training processes. Subsequent developments like
GNNRec [ 187] and XSimGCL [ 222] advanced session-based and graph contrastive learning recommendations, respec-
tively, addressing specific challenges such as social influence integration and bias mitigation. Ensuring trustworthiness
in GNN-based RS requires enhancements in robustness, explainability, and fairness to ensure reliable recommendations
[224].
Practical Challenges Addressed GNNs can effectively address various practical challenges by modeling complex
relationships in data. In e-commerce, models like LightGCN, GC-MC, NGCF, and Graph-ICF enhance personalization,
scalability, and efficiency for rating, link, and item predictions. These models are capable of handling extensive
product catalogs and large user bases efficiently. In social networks, GNNs such as GNN-SoR and GraphRec improve
user interaction predictions, boosting content relevance and user engagement by understanding social dynamics
and user relationships. In healthcare and finance, GNNs like KGAT and Fi-GNN provide secure and interpretable
recommendations, ensuring data privacy and compliance with regulations. These systems have ability to address the
13
Page 14:
arXiv Template A P REPRINT
cold-start problem by incorporating user and item features from knowledge graphs, providing accurate recommendations
even with limited initial data. The practical applications and use of GNNs are further detailed in Table 6.
Table 6 presents the use of GNNs, their variants, the data being used, evaluation metrics, and applications. For detailed
evaluation criteria on scalability, interpretability, computational efficiency, and reproducibility, please refer to the
Appendix.
Table 6: Comprehensive Overview of Graph Neural Network Models across Various Metrics and Use Cases. This
table details each model’s Input features, Year of Publication, and Characteristics such as Scalability, Interpretability,
Efficiency, and Reproducibility (rated as High, Medium, or Low, the symbol ’-’ means no information available for
this). It also lists the Dataset Used, Evaluation Metrics, Model Accuracy (as per evaluation metric from the previous
column), Learning Task, and Application Field.
Model Year Input Data Scalability, Inter-
pretability, Effi-
ciency, Reproducibil-
ityDataset Evaluation
MetricsModel Accuracy Application
GCN[202] 2015 MovieLens High, Medium, High,
HighMovieLens MSE MovieLens: 0.5 e-commerce
GC-
MC[189]2017 MovieLens-
1M,
MovieLens-
10M, Flixster,
Douban, Yahoo
MusicHigh, Low, -, High MovieLens-
1M,
MovieLens-
10M, Flixster,
Douban, Yahoo
MusicRMSE MovieLens-1M:
0.832
MovieLens-10M:
0.777
Flixster: 0.941
Douban: 0.734
Yahoo Music: 20.5e-commerce
NGCF[208] 2019 Gowalla,
Yelp2018,
Amazon-books-, Low, Medium, High Gowalla,
Yelp2018,
Amazon-booksRecall,
NDCGGowalla:
0.1569/0.1327
Yelp2018: 0.0579/
0.0477
Amazon-books:
0.0337/0.0261e-commerce
Graph-
ICF[193]2022 MovieLens-
1M, Pinterest-
20, Yelp-, Low, Medium, High MovieLens-
1M, Pinterest-
20, YelpHR, NDCG,
MAPMovieLens-1M:
0.7425/0.4555/0.3721
Pinterest-20:
0.8987/0.5830/0.4873
Yelp:
0.7519/0.4856/0.4033e-commerce
GNN-
SoR[169]2020 Epinions, Yelp,
Flixster-, -, -, Low Epinions, Yelp,
FlixsterRMSE,
MAE,
NDGCEpinions:
0.880/0.791/0.792
Yelp:
0.820/0.871/0.687
Flixster:
0.863/0.859/0.594Social Net-
work RecSys,
e-commerce
GCM[188] 2022 Yelp-NC, Yelp-
OH, Amazon-
book-, Low, Medium, High Yelp-NC, Yelp-
OH, Amazon-
bookHR@10,
NDGC@10Yelp-NC:
0.1046/0.0557
Yelp-OH:
0.2648/0.1457
Amazon-book:
0.0968/0.0536e-commerce
GCF-
YA[170]2019 MovieLens-
1M,
MovieLens-
10M, Taobao-, Low, -, Low MovieLens-
1M,
MovieLens-
10M, TaobaoHR@10,
NDGC@10MovieLens-1M:
0.7818/0.4873
MovieLens-10M:
0.7642/0.4677
Taobao:
0.3662/0.2491e-commerce
DGSR[179] 2023 Beauty, Games,
CDs-, Low, -, High Beauty, Games,
CDsNDCG@10,
Hit@10Beauty: 52.4/35.9
Games: 75.57/55.7
CDs: 72.43/51.22e-commerce
GraphRec[ 191]2019 Ciao, Epinions -, Low, -, High Ciao, Epinions MAE, RMSE Ciao: 0.7387/0.9794
Epinions:
0.8441/1.0878e-commerce
KGAT[197] 2019 Amazon-book,
Last-FM,
Yelp2018-, High, High, High Amazon-book,
Last-FM,
Yelp2018Recall@20
NDCG@20Amazon-book:
0.1489/0.1006
Last-
FM:0.0870/0.1325
Yelp2018:
0.0712/0.0867e-commerce
7.2 Sequential and Session-based Recommender Systems
Traditional models like Markov chains [ 306], pattern/rule mining [ 225], and latent factorization techniques [ 307] have
been long used in analyzing sequential data and user-item relationships by examining transitions, patterns, and latent
connections. However, they often struggle with dynamically predicting user preferences, typically due to a narrow
14
Page 15:
arXiv Template A P REPRINT
focus on immediate past users’ interactions or statistical correlations. This limitation is overcome by sequential RS
[95], which exploit the temporal order and context of user interactions. The evolution of sequential RS has transitioned
from Markov Chains and session-based KNN to sophisticated deep learning approaches, including RNNs, LSTMs,
attention mechanisms, and transformer architectures.
Sequential recommendation is commonly viewed as a next-item or next-basket prediction challenge [ 37]. Both the
sequential and session-based RS leverage user action sequences to anticipate users’ future preferences [ 95]. Specifically,
sequential RS consider the interaction histories of the users to predict future behaviour or users’ preferences. In contrast,
session-based RS, detailed in survey [ 38], focus on short-term user activity for real-time recommendations. These
approaches collectively enhance personalization and relevance across diverse platforms.
A sequential RS model can be defined as:
inext=f(history (u)),
where inextis the next recommended item, history (u)is the user u’s interaction sequence, and fmodels sequential
behavior to predict future interactions.
A session-based RS model can be defined as:
isession-next =g(scurrent),
withisession-next as the imminent session recommendation, scurrent representing the ongoing session interactions, and g
predicting the next item considering the session’s context.
The evolution of sequential and session-based RS has seen significant advancements with various models. For example,
Translation-based RS (TransRec) [ 308], integrates third-order interactions to enhance sequential predictions. The
research has progressed to using RNNs with GRU4Rec [ 158] and its enhancement, GRU4Rec+ [ 159], improving
session-based recommendations through refined loss functions and sampling strategies.
Subsequently, CNNs are applied in models like Convolutional Sequence Embedding Recommendation Model (Caser)
[245] and NextItNet [ 130], targeting effective session-based recommendations. The introduction of self-attention
mechanisms in Self-Attention based RS (SASRec) [ 161] for sequential model, and the exploration of item transitions
with Session-based Recommendations with Graph Neural Networks (SR-GNN) [216], showed further progress.
Recent developments have seen the application of the Transformers architecture, with models like Bert for RS
(BERT4Rec) [ 228] using bidirectional self-attention for deep sequence analysis, and Transformers4Rec [ 247] adapting
NLP transformers for recommendation contexts.
GNNs have been employed for modeling session-based interactions in GRASER [ 219], and LightSANs [ 139] improved
traditional Self-Attention Networks (SANs) by reducing complexity and refining sequence modeling with low-rank
decomposed self-attention.
Frequency Enhanced Hybrid Attention Network (FEARec) [ 240] and Knowledge Prompt-tuning for Sequential
Recommendation (KP4SR) [ 241] advance sequential recommendation by leveraging hybrid attention mechanisms and
integrating external knowledge bases, respectively, for better model performance.
Practical Challenges Addressed Sequential and session-based RS effectively tackle practical challenges by capturing
temporal dynamics and sequential patterns in user behavior. Models like TransRec, GRU4Rec, and GRU4Rec+ use
recurrent neural networks to ensure scalability and computational efficiency, making them ideal for e-commerce
and video streaming. Caser and NextItNet enhance these capabilities with convolutional layers, improving accuracy.
SASRec and SR-GNN apply self-attention mechanisms and graph neural networks to capture complex user-item
interactions in e-commerce and video games. BERT4Rec and Transformers4Rec leverage transformer architectures to
model long-range dependencies, achieving high accuracy across datasets like Amazon Beauty, Steam, and MovieLens.
These models also have the ability to address data sparsity and cold start issues by considering both short-term
session-based and long-term sequential preferences. They adapt to rapidly changing user interests and provide real-time
recommendations, making them effective in industries like entertainment and news. For instance, in the news industry,
they offer timely and relevant articles . In e-commerce, they track user interactions within a session to provide context-
aware product suggestions, enhancing the shopping experience and increasing the likelihood of immediate purchases.
The practical applications and use of these systems are further detailed in Table 7.
15
Page 16:
arXiv Template A P REPRINT
Table 7: Sequential Models. This table provides a detailed overview of various sequential models in recommendation
systems, showcasing their combined characteristics of Scalability, Interpretability, Computational Efficiency, and
Reproducibility (rated as High, Medium, or Low). Additionally, the table includes information on datasets used,
evaluation metrics, model accuracy, publication year, and application fields.
Model Year Input Data Scalability, Inter-
pretability, Efficiency,
ReproducibilityDataset Evaluation
MetricsModel Accuracy Application
TransRec
[308]2017 User-item inter-
action; Sequen-
tial behaviorHigh, Medium, High,
Yes3Epinions; Automo-
tive; Google Local;
Office Products;
Toys and Games;
Clothing, Shoes,
and Jewelry; Cell
Phones and Ac-
cessories; Video
Games; Electronics;
Foursquare; FlixterAUC; Hit@50 Epinions: 0.6133,
4.63%; Automotive:
0.6868, 5.37%;
Google Local:
0.8691, 6.84%;
Office Products:
0.7302, 6.51%; Toys
and Games: 0.7590,
5.44%; Clothing,
Shoes, and Jewelry:
0.7243, 2.12%;
Cell Phones and
Accessories: 0.8104,
9.54%; Video
Games: 0.8815,
16.44%; Electronics:
0.8484, 5.19%;
Foursquare: 0.9651,
67.09%; Flixter:
0.9750, 35.02%E-commerce,
Video Stream-
ing, Social
Media
GRU4Rec
[158]2016 User-item
graphs; node
featuresHigh , Medium , High ,
Yes4RecSys Challenge
20155; Youtube-
like OTT video
service platform.Recall@20;
MRR@20Item-KNN for:
RSC15: 0.5065,
0.2048; VIDEO:
0.5508, 0.3381E-commerce;
Video Stream-
ing
GRU4Rec+
[159]2018 Session-based;
RNN; GRUHigh, Medium, High,
Yes6RSC15; VIDEO;
VIDXL; CLASSRecall@20;
MRR@20RSC15: 0.7208,
0.3137; VIDEO:
0.6400, 0.3079;
VIDXL: 0.8028,
0.5031; CLASS:
0.3137, 0.1167E-commerce;
Video Stream-
ing; Classi-
fieds
Caser [245] 2018 Sequential;
CNNHigh, Medium, High,
Yes7MovieLens;
Gowalla;
Foursquare; TmallPrecision@N;
Recall@N;
MAPMovieLens: 0.2502,
0.0632, 0.1507;
Gowalla: 0.1961,
0.0845, 0.0928;
Foursquare: 0.1351,
0.1035, 0.0909;
Tmall: 0.0312,
0.0366, 0.0310Various
domains
NextItNet
[130]2019 Sequential;
CNN; Dilated
convolutionHigh, Medium, High,
Yes8Yoochoose-buys;
Last.fmMRR@20;
HR@20;
NDCG@20Yoochoose-buys:
0.1901, 0.4645,
0.2519; Last.fm:
0.3223, 0.4626,
0.3542E-commerce;
Music
SASRec
[161]2018 User-item
graphs; node
featuresHigh, Medium , High ,
Yes reproducibility9Amazon - Beauty;
Amazon - Games;
Steam; MovieLens-
1MHit@10;
NDCG@10Amazon - Beauty:
0.4854, 0.3219;
Amazon - Games:
0.7410, 0.5360;
Steam: 0.8729,
0.6306; ML-1M:
0.8245, 0.5905E-
Commerce;
Video Games;
Movies
SR-GNN
[216]2019 User-item
graphs; node
featuresHigh, Medium , High ,
Yes10YOOCHOOSE 1/64:
; YOOCHOOSE 1/4;
DIGINETICAP@20;
MRR@20YOOCHOOSE 1/64:
0.7057, 0.3094;
YOOCHOOSE 1/4:
0.7136, 0.3189;
DIGINETICA:
0.5073, 0.1759E-Commerce
3https://sites.google.com/a/eng.ucsd.edu/ruining-he/
4https://github.com/hidasib/GRU4Rec
5https://recsys.acm.org/recsys15/challenge/
6https://github.com/hidasib/GRU4Rec
7https://github.com/graytowne/caser
8https://github.com/fajieyuan/NextItNet
9https://github.com/kang205/SASRec
10https://github.com/CRIPAC-DIG/SR-GNN
16
Page 17:
arXiv Template A P REPRINT
Table 7 – continued from previous page
Model Year Input Data Scalability, Inter-
pretability, Efficiency,
ReproducibilityDataset Evaluation
MetricsModel Accuracy Application
BERT4Rec
[228]2019 User-item
graphs; node
featuresHigh , Medium , High ,
NoAmazon Beauty;
Steam; MovieLens-
1m; MovieLens-
20mHR@10;
NDCG@10;
MRR11Amazon
Beauty: 0.1599,
0.1862, 0.1701;
Steam:0.4013,
0.2261, 0.1949;
ML-1m: 0.6970,
0.4818, 0.4254;
ML-20m: 0.7473,
0.5340, 0.4785E-commerce;
Movies
Transformers
4Rec [247]2021 User-item
graphs; node
featuresHigh , Medium , High
Yes12REES46; YOO-
CHOOSE; G1;
ADRESSANGCG@20;
HR@20REES46: 0.2542,
0.4858; YOO-
CHOOSE: 0.3776,
0.6384; G1: 0.3675,
0.6721; ADRESSA:
0.3912, 0.7488E-commerce,
News
GRASER
[219]2022 Session-based;
Graph Neu-
ral Networks;
Non-sequential
interactionsHigh, Medium, High,
Yes13Yoochoose; Diginet-
icaMRR@20;
P@20Yoochoose: 0.3497,
71.37; Diginetica:
0.2045, 53.45E-commerce
LightSANs
[139]2021 Sequential;
Low-rank
decomposed
self-attentionHigh, Medium, High,
Yes14Yelp; Books; ML-
1MHIT@10;
NDCG@10Yelp: 0.5480,
0.2890; Books:
0.8760, 0.4250;
ML-1M: 0.2284,
0.1145E-commerce;
Books;
Movies
FEARec
[240]2023 Sequential;
Frequency-
based self-
attentionHigh, Medium, High,
Yes15Beauty; Clothing;
Sports; ML-1MHR@5;
HR@10;
NDCG@5;
NDCG@10Beauty: 0.0597,
0.0884, 0.0366,
0.0459; Clothing:
0.0214, 0.0323,
0.0121, 0.0156;
Sports: 0.0353,
0.0547, 0.0216,
0.0272; ML-1M:
0.2212, 0.3123,
0.1523, 0.1861E-commerce;
Movies
KP4SR [ 241] 2023 Sequential;
Knowledge
graph; Prompt-
tuningHigh, Medium, High,
Yes16Books; Music;
MoviesNDCG@5;
HR@5Books: 0.0609,
0.0824; Music:
0.0906, 0.1108;
Movies: 0.0755,
0.1058E-commerce;
Music;
Movies
7.3 Knowledge-based Recommender Systems
Knowledge Bases (KB), particularly Knowledge Graphs (KG), have been extensively used in the literature, for enhancing
personalized recommendations by leveraging user/item information [ 173]. A KG is a directed graph G= (V, E),
where VandErepresent entities and relations between them, respectively, with E⊆V×V. It includes entity type
function Φ :V→Aand relation type function Ψ :E→R, mapping entities to types Aand relations to types
R. KGs are depicted as sets of triples ⟨eh, r, et⟩, signifying a relation rfrom ehtoet. This relational information
helps RS understand user preferences and item relations, employing various methods to integrate KGs for improved
recommendations. KG-based RS can be observed through three primary approaches: Embedding-based, Path-based,
and Propagation-based approaches, each advancing the way RS leverage the rich relational data within KGs, as classified
by [173].
Embedding-based approaches focus on learning and applying embeddings to represent KG entities (nodes) and relations
(edges), enhancing user and item representations. They typically start with initial embedding generation using models
like TransE [ 309], TransD [ 198], and node2vec [ 310], followed by their application in RS through attention mechanisms
in KSR [229] or generative models like BEM [176] and KTGAN [311].
11HR@1, HR@5, NDCG@5 metrics dropped for simplicity.
12https://paperswithcode.com/paper/behavior-sequence-transformer-for-e-commerce
13https://github.com/tgdabe/GRASER
14https://github.com/RUCAIBox/LightSANs
15https://github.com/sudaada/FEARec
16https://github.com/zhaijianyang/KP4SR
17
Page 18:
arXiv Template A P REPRINT
Joint Learning Methods optimize both KG embeddings and recommendation components simultaneously using a unified
loss function. Examples include CKE [ 112], which integrates auto-encoders for item representations, and SHINE [ 312],
which acquires user embeddings from heterogeneous graphs. Multi-Task Methods such as KTUP [ 221] and MKR [ 206]
address KG-enhanced recommendation and KG completion concurrently, improving both entity/relation representations
and recommendations.
Path-based approaches utilize KG connectivity patterns. Meta-Structure-based Methods like KGCN [ 111] maintain
entity proximity in the latent space using graph convolution. Path-Embedding-based Methods, such as MCRec [ 313]
and RKGE [ 157], derive preference scores from path embeddings, incorporating meta-path information and RNN-based
path semantics.
Propagation-based approaches influence embeddings through multi-hop neighbor interactions within the KG. Item
KG-based methods like Ripplenet [ 215] aggregate item-related embeddings to derive user interests, whereas User-Item
KG-based methods such as KGAT [ 197] and Intentgc [ 195] refine both user and item embeddings by propagating
embeddings across a user-item graph, enhancing recommendation accuracy.
Practical Challenges Addressed KGs have become increasingly instrumental across various industries, leveraging
complex and rich datasets to build RS. For instance, in e-commerce, methods like TransE [ 309] and Node2Vec [ 310]
have been used to accurately suggest products by understanding the underlying connections between items and user
preferences. Similarly, in the movie recommendation space, models like KSR [ 229] and KTUP [ 221] utilize user-item
interactions and entity graphs to provide personalized movie suggestions. Social network platforms benefit as well,
with systems like SHINE [ 312] analyzing sentiment and social networks to enhance user engagement. Overall, these
systems enable more contextually aware, personalized, and efficient recommendation systems, significantly improving
user experience across these sectors. More details in Table 8.
Table 8: Comprehensive Overview of Knowledge Graph Based Recommender System Models across Various Metrics
and Use Cases. This table details each model’s Input features, Year of Publication, and Characteristics such as
Scalability, Interpretability, Efficiency, and Reproducibility (rated as High, Medium, or Low). It also lists the Dataset
Used, Evaluation Metrics, Model Accuracy, Learning Task, and Application Field.
Model Year Input data Scalability, Inter-
pretability, Compu-
tational Efficiency,
ReproducibilityDataset Evaluation
MetricsModel Accuracy Application
TransE[309] 2013 Item-item graph,
Multi-relational
relationshipsHigh, Medium, -, High Wordnet, Free-
base15k, Free-
base1MMean Rank,
Hits@10
(Raw/filtered)Wordnet: 263/251,
Freebase15k:
75.4/89.2,
Freebase1M:
243/125, 34.9/47.1,
14615/34.0Social net-
work analysis
Hete-
MF[111]2013 User-item
interaction,
Entity-relation
graph-, -, High, Low IMDb-MovieLens-
100KMAE, RMSE 0.778/0.9905 Movie recsys
HeteRec-
p[314]2014 User-item inter-
action, Implicit
feedbackLow, -, Low, Low IMDb-MovieLens-
100K, YelpPrecision@1,
MRR0.2121/0.0213 Movie recsys
Hete-
CF[315]2014 User-item rela-
tionship-, Low, High, High DBLP, Meetup MAE, RMSE DBLP: 0.856/0.994,
Meetup:
0.876/0.978Social Net-
work Recsys
TransD[198] 2015 Entity-relation
tripetsLow, Low, -, High Wordnet18, Free-
base 15kMean Rank
(raw and
filtered),
Hits@10 (raw
and filtered)Wordnet18:
224/212, 79.6/92.2,
Freebase 15k:
194/91, 53.4/77.3AI Applica-
tions
SemRec
[316]2015 User-item inter-
action-, High, Low, High Douban, Yelp RMSE, MAE Douban:
0.7844/0.6054,
Yelp: 1.2025/0.8901Movie,
Restaurant
Recsys, User
characteris-
tics analysis
and Recom-
mendation
explanation
Node2Vec[ 310]2016 Item-item graph High, Medium, High,
HighBlogCatalog, PPI,
Wikipedia, Face-
book, PPI, arXivMacro F1
score, AUCBlogCatalog(F1):
22.3, PPI(F1): 1.3,
Wikipedia(F1):
1.8, Face-
book(AUC): 0.9680,
PPI(AUC): 0.7719,
arXiv(AUC): 0.9366Data mining
18
Page 19:
arXiv Template A P REPRINT
Table 8 – continued from previous page
Model Year Input data Scalability, Inter-
pretability, Compu-
tational Efficiency,
ReproducibilityDataset Evaluation
MetricsModel Accuracy Application
KSR[229] 2018 User-item inter-
action sequence,
Entity graph-, High, -, High Last.FM, Ml-20M,
ML-1M, Amazon-
bookMAP, MRR,
Hit@10,
NDCG@10Last.FM:
0.427/0.427/
0.607/0.460,
Ml-20M: 0.294/
0.294/0.571/
0.344, Ml-1M:
0.356/0.356/
0.655/0.417,
Amazon-book:
0.353/0.353/
0.653/0.413e-commerce
KTGAN
[311]2018 User-movie in-
teraction-, Low, -, High Douban Precision@3,
Average Pre-
cision@3,
NDCG@30.759/0.701/ 0.771 Movie Rec-
sys
SHINE[312] 2018 Sentiment/
social/ profile
network-, -, -, High Weibo-STC, Wiki-
RfAAccuracy,
Micro-F1,
precision@K,
recall@KWeibo-STC:
0.855/0.881Social net-
work analysis
RippleNet
[215]2018 User-item inter-
action-, High, -, High MovieLens-1M,
Book-Crossing,
Bing-NewsAUC, Accu-
racyMovieLens-1M:
0.921/0.844, Book-
Crossing: 0.729/
0.662, Bing-News:
0.678/0.632e-commerce
BEM[176] 2019 Entity graph,
User interaction
graphLow, High, Medium,
HighFB15K237(KG) Hit@10 FB15K237(KG)+
pagelink: 44.72,
FB15K237(KG)+
desc: 44.58e-commerce
KTUP[221] 2019 User-item inter-
action-, High, -, High MovieLens-1m, DB-
book2014Precision@10,
Recall@10,
F1@10,
Hit@10,
NDCG@10
and Hit@10,
MeanMovieLens-1m:
41.03/17.25/
19.82/89.03/ 69.92,
DBbook2014:
4.05/24.51/
6.73/34.61/ 27.62Movie Rec-
sys
MKR[206] 2019 User-item in-
teraction, KG
triples-, Medium, -, High MovieLens-1M,
Book-Crossing,
Last.FM, Bing-
NewsAUC, ACC,
RMSEMovieLens-1M:
0.917/0.843/ 0.302,
Book-Crossing:
0.734/0.704/
0.558, Last.FM:
0.797/0.752/
0.471, Bing-News:
0.689/0.645/ 0.459e-commerce,
News
RCF[317] 2019 Item relations,
User-item
interactionMedium, High, -, High MovieLens, KKBox HR@20,
MRR@20,
NDCG@20MovieLens:
0.2354/0.0642/
0.1015, KKBox:
0.8563/0.5762/
0.6412e-commerce
Akupm[318] 2019 User-item
implicit in-
teraction,
Entity-relation
graph-, -, -, High MovieLens-1, Book-
CrossingAUC, ACC MovieLens-1:
0.918/0.845,
Book-Crossing:
0.843/0.807e-commerce
KNI[319] 2019 User-item inter-
action, Knowl-
edge graph-, Medium, -, High C-Book, Movie-1M,
A-Book, Movie-
20MAUC, Accu-
racyC-Book:
0.7723/0.7063,
Movie-1M:
0.9449/0.8721,
A-Book:
0.9238/0.8472,
Movie-20M:
0.9704/0.9120e-commerce
IntentGC
[195]2019 User-item Ex-
plicit interactionHigh, -, -, High Taobao, Amazon AUC, MRR Taobao:
0.701740/0.3746,
Amazon:
0.837589/2.7981e-commerce
19
Page 20:
arXiv Template A P REPRINT
Table 8 – continued from previous page
Model Year Input data Scalability, Inter-
pretability, Compu-
tational Efficiency,
ReproducibilityDataset Evaluation
MetricsModel Accuracy Application
PGPR[213] 2019 User-item inter-
action, Item fea-
tures-, High, -, High CDs & Vinyl, Cloth-
ing, Cell Phones,
BeautyNDCG, Recall,
HR, Prec.CDs & Vinyl:
5.590/7.569/
16.886/2.157, Cloth-
ing: 2.858/4.834/
7.020/0.728,
Cell Phones:
5.042/8.416/
11.904/1.274,
Beauty:
5.449/8.324/
14.401/1.707e-commerce
KGSF[320] 2020 User-item in-
teraction, Node
features-, High, -, High REDIAL Recall@k (k =
1, 10, 50)0.039/0.183/ 0.378 e-commerce
KIM[321] 2021 User-item inter-
action, Entity
graph-, High, -, High MIND, Feeds AUC, MRR,
nDCG@5,
nDCG@10MIND: 67.13 ±0.29/
32.08 ±0.24/
35.49 ±0.34/
41.79 ±0.28, Feeds:
66.45 ±0.13/
30.27 ±0.09/
35.04 ±0.09/
40.43±0.12Online News
Recsys
BCIE[322] 2023 User-item inter-
action, Item fea-
tures-, -, -, High MovieLens, Ama-
zonBookNARC,
Hits@kMovielens 20M:
0.185/0.192 Ama-
zonBook: 0.18/
0.205e-commerce
DiffKG[323] 2024 User-item inter-
action, Item fea-
tures-, High, -, High Last-FM, MIND,
Alibaba-iFashionMetrics Last-FM:
0.0980/0.0911,
MIND:
0.0615/0.0389,
Alibaba-iFashion:
0.1234/0.0773e-commerce
7.4 Reinforcement Learning-based Recommender Systems
Reinforcement learning (RL) [ 277] is a subset of ML where an agent learns to make decisions by interacting with
an environment, aiming to achieve a goal through trial and error, guided by rewards for its actions, without explicit
instructions on what actions to take. Deep Reinforcement Learning-based methods [ 256] integrate RL with deep neural
networks to enable agents to handle complex modalities of the data directly. Given a set of states S, a set of actions A, a
reward function R, a transition probability function P, and a discount factor γ, the goal of the RL agent is to find a
policy πthat maximizes the expected, discounted cumulative reward over time. The mathematical formulation is [ 43]:
max
πE"TX
t=0γtr(st, at)#
, (8)
where tindexes the time steps, ranging from 0 to T, the maximum time step in a finite Markov Decision Process (MDP),
standatrepresent the state and action at time t, respectively, r(st, at)is the immediate reward received after taking
action atin state st,γtapplies the discount factor to future rewards, making them worth less than immediate rewards.
Applying these RL concepts to RS, the RS itself acts as the RL agent [ 254] through an environment constituted by user
interactions and data, as detailed in a related survey [43].
RL methods in RS has evolved into two primary frameworks: traditional RL-based RS and deep learning-enhanced
RL-based RS. Traditional methods, such as Q-learning [ 255] and SARSA [ 324], optimize policies within Markov
Decision Processes (MDP) using model-free approaches, with applications well-documented across various contexts
[325,269,271,267,257]. These methods often leverage Monte Carlo Tree Search (MCTS) [ 326] for effective
simulation and policy refinement.
Deep learning methods in RL-based RS [ 43], on the other hand, incorporate advanced neural network architectures to
enhance policy learning. These include Vanilla Deep Q-Network (DQN) and its variants [ 255,259,253,131,214,264,
258], which utilize neural networks for accurate action-reward estimation. Hybrid methods like Actor-Critic and Deep
Deterministic Policy Gradient (DDPG) [ 253,327,281,252], and Soft Actor-Critic (SAC) [ 328,262,266] blend value
and policy strategies to balance exploration and exploitation effectively.
20
Page 21:
arXiv Template A P REPRINT
Furthermore, model-based RL approaches in RS focus on simulating user behavior to tailor recommendations, with
techniques ranging from generative adversarial networks [ 119] to multi-agent systems [ 260,268]. These sophisticated
methods aim to predict user interactions and refine recommendations continually, enhancing personalization and
contextual relevance in RS.
Practical Challenges Addressed RL is being used in RS to improve personalization problem . For instance, in
e-commerce, RL enhances personalization and improves customer satisfaction by continuously learning from user
interactions to optimize recommendation strategies, as shown in systems used by Amazon and Taobao [ 212,262]. In
the media sector, RL aids in curating more engaging content recommendations, like music and news, by analyzing
sequential interaction data to predict future preferences [ 267,259]. Additionally, in job recommendation systems,
RL algorithms optimize outcomes by suggesting roles that align closely with the users’ evolving career interests and
skills [ 257]. By employing techniques such as deep Q-networks and policy gradient methods, RL-based recommender
systems continuously refine their decision-making processes, leading to improved long-term user engagement and
satisfaction. More details in Table 9.
Table 9: Comprehensive Overview of Reinforcement Learning based Recommender System Models across Various
Metrics and Use Cases. This table details each model’s Input features, Year of Publication, and Characteristics such as
Scalability, Interpretability, Efficiency, and Reproducibility (rated as High, Medium, or Low). It also lists the Dataset
Used, Evaluation Metrics, Model Accuracy, Learning Task, and Application Field. Metrics that their numerical value is
not reported are specified with “No numerical value”.
Model Year Input data Scalability, Inter-
pretability, Compu-
tational Efficiency,
ReproducibilityDataset Evaluation
MetricsModel Accuracy Application
RLWRec
[269]2017 User-item inter-
actions-, -, -, Low Low, medium, Large
Music datasetAccuracy,
Score, Cover-
ageNo numerical value Music rec
DAHCR
[223]2023 User-item
graphs; Node
features-, -, -, High LastFM*, Yelp* Success Rate,
Average Turns,
hDCG@(T, K)LastFM: 0.925/6.31/
0.431
Yelp*: 0.626/11.02/
0.192e-commerce
LIRD[252] 2017 User-item
graphs; node
features-, -, -, Low E-commerce web-
siteMAP, NDGC No numerical value e-commerce
Multi
With[260]2017 User-item
graphs; node &
item features-, -, -, Low ACM data set MRR, P@3,
P@5, P@10,
NDCG@3,
NDCG@5,
NDCG@100.601/0.437/
0.321/0.178/
0.561/0.560/ 0.565Author Rec-
sys
[272] 2018 User-item inter-
actions-, -, -, Low Data logs from e-
learningRMSE 0.71 e-learning
DRN[259] 2018 User-item inter-
actions; node
features-, -, -, Low News recommenda-
tionsOffline: CTR,
NDCG; On-
line: CTR,
Precision@5,
nDCGOffline:
0.1662/0.487
Online:
0.0113/0.0149/
0.0492News recom-
mendation
DeepPage
[253]2018 User-item
sessions-,-, -, Low E-commerce data Offline: Pre-
cision@20,
Recall@20,
F1-score@20,
NDCG@20,
MAP0.0491/0.3576/
0.0805/0.1872/
0.1378e-commerce
[271] 2018 User-item
graphs; node
features-, Low, Medium, Low Movielens-100k,
Movielens-1MP@30, R@30 Movielens-100k:
0.246/0.169
Movielens-1M:
0.277/0.155e-commerce
[119] 2018 User-item
graphs; node
features-, Medium, -, High MovieLens, LastFM,
Yelp, Taobao, Yoo-
Choose, Ant Finan-
cialReward, CTR Combined datset:
25.36/0.7817e-commerce
SADQN
[131]2019 User-item
graphs; user-
user graph-, -, -, Low LastFM, Ciao, Epin-
ionsHR,
NDCG@10HR: 0.5438 ±0.0036/
0.4256 ±0.0031/
0.4755±0.0016
NDCG@10: No nu-
merical valuee-commerce
CROMA
[264]2019 User-item
graphs; node
features-, -, -, High Twitter Precision, Re-
call, F-Score,
MRR, Hits@574.55/74.09/
74.32/81.85/ 95.00Social net-
work recom-
mendations
DRCGR
[258]2019 User-item
graphs-, -, -, Low E-commerce dataset MAP, NDCG No numerical value e-commerce
17Shown only the best results
21
Page 22:
arXiv Template A P REPRINT
Table 9 – continued from previous page
Model Year Input data Scalability, Inter-
pretability, Compu-
tational Efficiency,
ReproducibilityDataset Evaluation
MetricsModel Accuracy Application
PGPR[213] 2019 User-item
graphs; node
features-, -, -, High CDs & Vinyl, Cloth-
ing, Cell Phones,
BeautyNDCG@10,
Recall@10,
HR@10,
Prec@10CDs & Vinyl:
5.590/7.569/
16.886/2.157
Clothing:
2.858/4.834/
7.020/0.728
Cell Phones:
5.042/8.416/
11.904/1.274
Beauty:
5.449/8.324/
14.401/1.707e-commerce
PGCR[329] 2019 User-item
graphs; node
features-, -, -, Low Music recommenda-
tion (KKBox)Average
rewardAccuracy Music recom-
mendation
SLATEQ
[273]2019 User-item
graphs; node &
item featuresHigh, -, -, - - - - Music Recsys
GCQN[214] 2020 User-item
graphs; node
features-, -, -, Low LastFM, ML1M,
PinterestMean of re-
wards received
in a T-step
episodeLastFM: 0.404
ML1M: 0.658
Pinterest: 0.215e-commerce
MASSA
[262]2020 User-item
graphs; node
features-, -, -, Low Taobao Precision,
nDCG0.615, 0.516 e-commerce
KERL[230] 2020 User-item
graphs; node
features, KG-, -, -, Low Beauty, CD, Books,
LastFMHR@10,
NDCG@10Beauty: 54.1/36.5
CD: 73.7/50.8
Books: 80.0/57.1
LastFM: 64.2/50.1
18e-commerce
KGPolicy
[212]2020 User-item
graphs; KG-, -, Medium, High Amazon-book, Last-
FM, Yelp2018Recall@20,
NDCG@20Amazon-book:
0.1572/0.1089
Last-FM:
0.0932/0.1472
Yelp2018:
0.0747/0.0921e-commerce
BatchRL-
MTF[266]2022 User-item
graphs; node
featuresHigh, -, -, Low Short video recom-
mendationOffline: Long-
term user satis-
faction per ses-
sion; Online:
App dwell time,
User positive-
interaction rateOffline: 4.126
Online: +2.550%
/+9.651%e-commerce,
video
TRIGR[217] 2022 User-item
graphs; node &
item featuresMedium, -, Medium,
HighMusic, Beauty,
ClothingHR@10,
F1@10,
NDGC@10Music:
0.9886/0.2304/
0.9436
Beauty:
0.8845/0.1798/
0.6949
Clothing:
0.7544/0.1405/
0.4865e-commerce
UCSRDRL
[278]2021 User-item
graphs; node
features-, -, -, Low Item-info, Trainset
and Track2_testsetModel score FUXI AI Lab Test
data: 1033481948e-commerce
RPMRS
[267]2021 User-item inter-
action logs-, -, -, Low Music & User logs Avg. score No numerical value e-commerce
MDP[257] 2021 User-item inter-
actions-, -, -, Low Transactional data of
job applications% improve-
ment of wage,
market revenue,
worker success
measures22%, 1.5-6%, 4x e-learning
7.5 Large Language Model based Recommender Systems
Language is a fundamental tool for human communication, essential for expressing thoughts, feelings, and intentions.
The challenge of understanding and leveraging human language has been a central pursuit in NLP research, leading
to significant developments in language modeling [ 282]. Early statistical models relied on the Markov assumption
18Shown next-item recommendation metrics
22
Page 23:
arXiv Template A P REPRINT
to predict word sequences [ 330,331], while subsequent neural language models utilized neural networks to estimate
probabilities of word sequences [ 332,333,334]. The advent of pre-trained language models like BERT and others
[233,335,336] marked a pivotal advancement, providing deep contextual insights that greatly enhanced NLP tasks.
The Transformer architecture and its attention mechanism allow for the efficient handling of long-range dependencies
and context [ 337]. The scaling laws suggest that larger models and datasets generally yield better performance [ 336],
leading to the development of Large Language Models (LLMs) [ 294], which demonstrate sophisticated capabilities in
AI tasks such as in-context learning and commonsense reasoning [ 338]. The integration of LLMs into RS [ 287,289]
has prompted extensive research and ongoing innovation, with comprehensive reviews and analyses provided by recent
surveys [51, 50, 284, 290], outlining the evolving landscape of LLM-based RS technologies.
The integration of BERT-like models into RS has led to significant advancements. Initial applications like BERT4REC
[228] utilized deep bidirectional self-attention for modeling user behavior sequences, while further developments
employed BERT for tasks ranging from conversational RS [ 78] to CTR prediction [ 339]. Enhancements in BERT-based
models have addressed specific RS challenges, such as item alignment in dialogues [ 340] and user representation
through models like U-BERT [ 341] and UserBERT [ 342]. Further innovations include BERT-based re-ranking [ 343]
and addressing data sparsity in group recommendations [344].
Prompt-based and in-context learning (ICL) approaches have leveraged the adaptability of LLMs, employing person-
alized prompts and natural language processing to enhance recommendation relevance and user interaction without
extensive retraining [ 345,293]. These methods have proven effective in various scenarios, from news recommendation
[346] to conversational and zero-shot recommendations, addressing longstanding issues like cold starts and data sparsity
[287, 288].
Moreover, advancements in prompt tuning and personalized recommendation strategies demonstrate the ongoing
evolution of LLM applications in RS, significantly improving system performance while also highlighting challenges
such as ethical considerations and the management of popularity biases [ 286,283]. These developments indicate a
move towards more sophisticated, context-aware systems that can dynamically adapt to user preferences and behaviors.
Practical Challenges Addressed LLMs have advanced RS by addressing key challenges such as the cold-start
problem, enhancing personalization, and improving accuracy. Models like BERT4REC [ 228] and UserBERT [ 342],
GBERT [ 344] and RecMind [ 293] effectively utilize user and item metadata to generate relevant suggestions in e-
commernce and entertainment. These models also support dynamic learning, allowing systems to adapt based on
real-time interactions, thus enhancing user engagement and satisfaction.
Table 10: Comprehensive Overview of LLM Based Models across Various Metrics and Use Cases. This table details
each model’s Input features, Year of Publication, and Characteristics such as Scalability, Interpretability, Efficiency, and
Reproducibility (rated as High, Medium, or Low). It also lists the Dataset Used, Evaluation Metrics, Model Accuracy,
Learning Task, and Application Field.
Model Year Input Data Scalability, Inter-
pretability, Compu-
tational Efficiency,
ReproducibilityDataset Evaluation
MetricsModel Accuracy Application
BERT4REC
[228]2019 User-item
graphs; node
featuresHigh, Medium, High,
High19Amazon Beauty,
Steam, MovieLens-
1m, MoveiLens-
20m)HR@10,
NDCG@10,
MRR20Beauty: 0.3025,
0.1862, 0.1701;
Steam: 0.4013,
0.2261, 0.1949;
ML-1m: 0.6970,
0.4818, 0.4254;
ML-20m: 0.7473,
0.5340, 0.4785E-commerce,
Video Games,
Movies
CTR-BERT
[339]2021 User-item
graphs; node
featuresHigh, Medium, High,
LowCurated CTR AUC Delta OOD: +2.27%, ID:
+2.17%Marketing /
CTR
MESE[340] 2022 User-item
graphs; node
featuresHigh, Medium, High,
High21ReDial, INSPIRED;
both from AMTR@1, R@10,
R@50ReDial: 5.6, 25.6,
45.5; INSPIRED:
4.8, 13.5, 30.1Movies
U-BERT
[341]2021 User-item
graphs; node
featuresHigh, Medium High,
LowAmazon (Office,
Video, Music,
Toys, Kindle), Yelp
ChallengeMSE Office: 0.6774;
Video: 0.8750;
Music: 0.7723;
Toys: 0.7823; Kin-
dle: 0.5912; Yelp:
1.5907E-Commerce
19https://github.com/FeiSun/BERT4Rec
20HR@1, HR@5, NDCG@5 are dropped for convenience.
21https://github.com/by2299/MESE
23
Page 24:
arXiv Template A P REPRINT
Table 10 – continued from previous page
Model Year Input Data Scalability, Inter-
pretability, Compu-
tational Efficiency,
ReproducibilityDataset Evaluation
MetricsModel Accuracy Application
UserBERT
[342]2022 User-item
graphs; node
featuresHigh, Medium,High,
High (Unofficial22)News, CTR For News:
AUC,
nDCG@10;
For CTR: AU,
APNews: 62.87 ±0.14,
40.64 ±0.12;
CTR: 73.96 ±0.06,
76.72±0.06News, Mar-
keting /
CTR
BECR [343] 2022 User-item
graphs; node
featuresHigh, Medium, High,
High23Trained on Ro-
bust04, ClueWeb09-
Cat-B; Evaluated on
MS MARCO dev
set, TREC DL19
and TREC DL20For Perfor-
mance24:
Training:
NDCG@20,
P@20, MS-
MARCO:
MRR@10Dev;
DL19 and
DL20:
NDCG@10Training on Ro-
bust04: 0.4656,
0.4005; Training on
ClueWeb09-Cat-B:
0.3075, 0.3987;
Evaluation on DL19:
0.658; Evaluation
on DL20: 0.647;
Evaluation on
MSMARCO: 0.319General:
text retrieval
research
MLPR [347] 2022 User-item
graphs; node
featuresHigh,Medium, High,
HighOne month data
from Walmart.comAUC;
NDCG@125Click: +6.48%,
+17.22%; ATC:
+4.66%, +10.61%;
Purchase: +1.03%,
+5.36%E-commerce
GBERT
[344]2022 User-item
graphs; node
featuresHigh, Medium, High,
LowWeeplaces; Yelp;
DoubanN@10; R@10
26Weeplaces: 36.43%,
52.82%; Yelp:
38.11%, 53.14%;
Douban: 54.58%,
79.90%Social
Networks;
Business
Reviews
Prompt4NR
[346]2023 User-item
graphs; node
featuresHigh, Medium, High,
High27MIND AUC; MRR;
NDCG@5;
NDCG@10Hybrid28: 69.64%,
34.26%, 38.30%,
44.33%News
P5 [345] 2022 User-item
graphs; node
featuresHigh,
Medium,High,High29Amazon (Sports,
Beauty, Toys), YelpFor Perfor-
mance on
Sequential
Recommen-
dations30:
HR@10,
NDCG@1031;For P5-B32Amazon
Sports: 0.0460,
0.0336; Amazon
Beauty: 0.0.0645,
0.0416; Amazon
Toys: 0.0675,
0.0536E-commerce
RecMind
[293]2024 User-item
graphs; node
featuresHigh, Medium,
High,LowAmazon Reviews -
Beauty; YelpFor Perfor-
mance on
Sequential
Recommen-
dations33:
HR@10,
NDCG@1034For RecMind-SI
(few-shot)35Ama-
zon Reviews -
Beauty: 0.1559,
0.1063; Yelp:
0.2451, 0.1607E-commerce;
Restaurants
RecRec
[348]2023 User-item
graphs; node
featuresHigh, Medium, High,
High36MovieLens-
100K; AliEC
Ads; GoodreadsSuccess Rate;
Number of
Changes
Required;
Side-effect on
User Recom-
mendationsMovieLens-100K:
100%; AliEC Ads:
>80%; Goodreads:
>90%Movies; Ads;
Books
22https://github.com/ilovemyminutes/UserBERT
23https://github.com/yingrui-yang/BECR
24The four other questions related to inference time, etc. are dropped.
25NDCG@5 was omitted.
26N@5 and R@5 are excluded for simplicity.
27https://github.com/resistzzz/Prompt4NR
28We dropped discrete and continuous templates for simplicity.
29https://github.com/jeykigung/P5
30We dropped other details about performance on rating, explanation generation and review preference, and considered only the
performance comparison on sequential recommendation because its the most relevant factor in this case.
31HR@5 and NDCG@5 are dropped for simplicity.
32Only P5-base scenario is considered for simplicity.
33We dropped other details about performance on rating, explanation generation and review preference, and considered only the
performance comparison on sequential recommendation because its the most relevant factor in this case.
34HR@5 and NDCG@5 are dropped for simplicity.
35Only RecMind-SI (few-shot) scenario is considered for simplicity and its high performance.
36https://github.com/hidasib/GRU4Rec
24
Page 25:
arXiv Template A P REPRINT
Table 10 – continued from previous page
Model Year Input Data Scalability, Inter-
pretability, Compu-
tational Efficiency,
ReproducibilityDataset Evaluation
MetricsModel Accuracy Application
TALLRec
[295]2023 User-item
graphs; node
featuresHigh, Medium, High,
High37MovieLens-100K;
BookCrossingAUC MovieLens-100K:
0.7198; BookCross-
ing: 0.6438Movies;
Books
GenRec
[286]2023 User-item
graphs; node
featuresHigh, Medium, High,
High38MovieLens 25M;
Amazon ToysHR@10,
NDCG@1039MovieLens 25M:
0.1311, 0.0837;
Amazon Toys:
0.0251, 0.0157Movies;
E-commerce
7.6 Multimodal Recommender Systems
Multimodality involves using and analyzing various data types—text, images, audio, video to enhance processing and
understanding. Multimodal RS utilize these diverse inputs to improve recommendation quality and user experience
by better understanding user preferences and item features [ 296]. These systems overcome the limitations of single-
modality systems through effective integration of heterogeneous data.
The evolution of multi-modal RS began with the introduction of Visual Bayesian Personalized Ranking (VBPR) [ 349],
which enhances personalized ranking by integrating visual features from product images. The results showed improved
accuracy and addressing cold-start issues. Attentive Collaborative Filtering (ACF) [ 350] introduced a novel attention
mechanism to better handle item- and component-level feedback in multimedia recommendations.
Further advancements were made with the development of the Multi-modal Knowledge Graphs (MKGs) [ 194],
a hybrid transformer with multi-level fusion, for tasks like link prediction and entity relation extraction. Online
Distillation-enhanced Multi-modal Transformer (ODMT) [ 244] uses diverse data types (ID, text, image) and an ID-
aware Multi-modal Transformer with online distillation to enhance feature interaction. These models showed substantial
performance increase in recommendation accuracy.
Collaborative Cross Networks (CoNet) [ 351] utilizes deep transfer learning. Multi-Modality enriched Sequential
Recommendation (MMSR) [ 249], a graph-based model, adaptively fuses multi-modal information to dynamically
prioritize modalities based on their sequential relationships. The Bayesian Multi-Modal recommendation Model
(BM3) [ 301] simplifies training by avoiding auxiliary graphs and negative samples with multi-modal data. The Multi-
modal Interest-aware Sequence Representation for Recommendation (MISSRec) [ 243] overcame the limitations of
ID-based models by leveraging multi-modal information for robust, generalizable sequence representations. Multi-
modal Recommendation (MMRec) [ 297], is another RS that provides a configurable platform for testing multimodal
recommendation models.
Multi-level Self-supervised Learning for Multimodal Recommendation (MENTOR) [ 298] employes multi-level self-
supervised tasks to improve model performance, though it required substantial computational resources. Recently,
Multi-modal Knowledge Distillation (PromptMM) [ 300] simplified the recommendation process through multi-modal
knowledge distillation and prompt-tuning.
Practical Challenges Addressed Multimodal RS are useful for e-commerce and social media platforms, where
diverse data sources and user interactions are prevalent. Models like VBPR, ACF, and CoNet are designed to be scalable
and computationally efficient, providing quick recommendations even with extensive user data. These models can
integrate various data types, such as text, images, and behavioral data, and can adapt to new trends and handle complex
user-item interactions.These RS improve personalization by leveraging the rich information from different modalities,
leading to more accurate and relevant recommendations. More details are provided in Table 11.
37https://github.com/SAI990323/TALLRec
38https://github.com/rutgerswiselab/GenRec
39HR@5 and NDCG@5 are dropped for simplicity.
25
Page 26:
arXiv Template A P REPRINT
Table 11: Comprehensive Overview of Multi-Modal Based Models across Various Metrics and Use Cases. This
table details each model’s Input features, Year of Publication, and Characteristics such as Scalability, Interpretability,
Efficiency, and Reproducibility (rated as High, Medium, or Low). It also lists the Dataset Used, Evaluation Metrics,
Model Accuracy, Learning Task, and Application Field.
Model Year Input Data Scalability, Inter-
pretability, Compu-
tational Efficiency,
ReproducibilityDataset Evaluation
MetricsModel Accuracy Application
VBPR [349] 2016 User-item
graphs; node
featuresHigh, Medium, High,
High40Amazon Women;
Amazon Men;
Amazon Phones;
Tradesy.comAUC 0.7834, 0.7841,
0.8052, 0.7829E-commerce
ACF [350] 2017 User-item
graphs; node
featuresHigh, Medium, High,
High41Pinterest; Vine HR@100;
NDCG@100Pinterest: 0.3378,
0.0855; Vine:
0.6365, 0.1903Images;
Videos
ODMT [ 244] 2023 User-item -, -, -, High42Stream; Arts; Office;
H&MRecall@10;
NDCG@10Stream: 0.1194,
0.0672; Arts:
0.1127, 0.0787;
Office: 0.1175,
0.0893; H&M:
0.1235, 0.0771Streaming
Media;
E-commerce
CoNet [351] 2018 User-item
graphs; node
featuresHigh, Medium, High,
High43Mobile Apps (Chee-
tah Mobile); Ama-
zon BooksHR@10;
NDCG@10Mobile Apps:
0.8480, 0.6887;
Amazon Books:
0.5338, 0.3424Apps; Books
MMSR [ 249] 2023 User-item -, -, -, High44Amazon Beauty;
Amazon Clothing;
Amazon Sports;
Amazon Toys;
Amazon Kitchen;
Amazon PhoneHR@5;
MRR@5Amazon Beauty:
7.1563, 4.4429;
Amazon Clothing:
1.8684, 1.1365;
Amazon Sports:
3.2657, 1.9846;
Amazon Toys:
6.1159, 3.8987;
Amazon Kitchen:
2.2145, 1.4238;
Amazon Phone:
6.9550, 3.9911E-commerce
BM3 [301] 2023 User-item -, -, -, High45Baby; Sports; Elec-
tronicsRecall@10;
NDCG@10Baby: 0.0564,
0.0301; Sports:
0.0656, 0.0355;
Electronics: 0.0437,
0.0247E-commerce
MISSRec
[243]2023 User-item -, -, -, High46Amazon Beauty;
Amazon Clothing;
Amazon SportsRecall@10;
NDCG@10Amazon Beauty:
0.0321, 0.0189;
Amazon Clothing:
0.0387, 0.0215;
Amazon Sports:
0.0268, 0.0159E-commerce
MMRec
[297]2023 User-item -, -, -, High47Amazon Review
DataN/A N/A E-commerce
MMSSL
[298]2023 User-item -, -, -, High48Amazon Baby;
TikTok; Allrecipes;
SportsRecall@20;
NDCG@20Amazon Baby:
0.0962, 0.0422;
TikTok: 0.0921,
0.0392; Allrecipes:
0.0367, 0.0135;
Sports: 0.0998,
0.0470Social Media;
E-commerce;
Cooking;
Sports
40https://github.com/example/VBPR
41https://github.com/example/ACF
42https://github.com/xyliugo/ODMT
43https://github.com/CoNetModel/CoNet
44https://github.com/HoldenHu/MMSR
45https://github.com/enoche/BM3
46https://github.com/gimpong/MM23-MISSRec
47https://github.com/enoche/MMRec
48https://github.com/HKUDS/MMSSL
26
Page 27:
arXiv Template A P REPRINT
Table 11 – continued from previous page
Model Year Input Data Scalability, Inter-
pretability, Compu-
tational Efficiency,
ReproducibilityDataset Evaluation
MetricsModel Accuracy Application
PromptMM
[300]2024 User-item
graphs; node
featuresHigh, Medium, High,
High49Netflix; TikTok;
ElectronicsRecall@20;
NDCG@2050Netflix: 0.1864,
0.0743; TikTok:
0.3054, 0.1013;
Electronics: 0.0737,
0.0258Video En-
tertainment,
Social Media
(Micro-
Video),
E-commerce
8 Specialized Recommender Systems
Specialized RS can be defined as those RS that are tailored to meet specific needs across various domains, using
advanced techniques to address unique user preferences or situations. Unlike general RS, these focus on specialized
techniques, functions and targeted recommendations. In the following subsections, we will explore these specialized
systems in detail.
8.1 Context-aware Recommender Systems
Context-aware recommender systems (CARS) are advanced RS that enhance the personalization of content by in-
corporating contextual information into the recommendation process [ 33]. Unlike traditional RS that primarily rely
on user-item interactions, CARS consider additional dimensions such as time, location, social settings, and user
behavior patterns to deliver more relevant and timely suggestions [ 352]. These systems have evolved to address specific
challenges such as the cold-start problem, where limited initial data is available about new users or items.
The core models employed in CARS span a variety of sophisticated algorithms designed to leverage contextual
information effectively into the recommendation process. Among these, factorization machines (FM) [ 353] are
prominent for their ability to capture interactions between variables within large datasets. Field-Aware Factorization
Machines (FFMs) [ 354] are specifically optimized for CTR prediction, showing the versatility and depth of models
developed for enhancing CARS’ performance. The Neural Factorization Machine (NFM) [ 98] extends FM by modeling
second-order feature interactions with the non-linearity of neural networks for higher-order interactions.
Deep learning has significantly advanced CARS by enabling sophisticated feature extraction and integration of diverse
data types, such as images and sequences [ 355]. Models like CNNs and LSTMs can process complex inputs and
temporal sequences, enhancing the system ability to understand and utilize context like time and location effectively.
DeepFM [ 99] merges FM recommendation capabilities with a novel neural network architecture. xDeepFM [ 141]
further extends the DeepFM concept by explicitly learning bounded-degree feature interactions while also capturing
arbitrary low- and high-order interactions implicitly. Additionally, scalability allows these models to maintain high
performance even with vast datasets, ensuring personalized recommendations.
Techniques such as attention mechanisms make recommendations more adaptive and context sensitive. The Attentional
Factorization Machine (AFM) [ 356] introduces a neural attention network to show the significance of each feature
interaction, enhancing model interpretability and efficiency. The Graph Convolution Machine (GCM) [ 188] and the
Attention-based Context-aware Sequential Recommendation model using Gated Recurrent Unit (ACA-GRU) [ 226]
both enhance RS by effectively synthesizing user, item, and context information into actionable insights.
Various CARS have been developed for different use cases. These include a tourism RS for personalized suggestions
[225], and a context-aware paper citation RS [ 134] that utilizes graph CNN combined with BERT for effective
document and context encoding. There are also CARS designed for smart product-service systems [ 357] and for cultural
heritage [ 66]. Moreover, the Sequential Model for Context-Aware Point of Interest (POI) Recommendation (SCR)
[237] enhances next POI predictions by integrating short-term preferences with multi-head attentive aggregation and
long-term preferences through context-aware layers.
Practical Challenges Addressed CARS enhance industries by efficiently handling diverse data sources and en-
suring scalability, interpretability, and computational efficiency. Models like FFM, NFM, and DeepFM are ideal for
e-commerce, advertising, and web platforms. They build user trust by making recommendations understandable and
49https://github.com/HKUDS/PromptMM
50Recall@50 and NDCG@50 are dropped for simplicity.
27
Page 28:
arXiv Template A P REPRINT
reproducible. CARS adapt to new data trends and manage complex interactions, providing personalized recommenda-
tions. Applied across various fields, including advertising, e-commerce, and social networks, these systems improve
operational efficiency and user satisfaction. More details in Table 12.
Table 12: Comprehensive Overview of Context-Aware Recommender System Models across Various Metrics and Use
Cases. This table details each model’s Input features, Year of Publication, and Characteristics such as Scalability,
Interpretability, Efficiency, and Reproducibility (rated as High, Medium, or Low). It also lists the Dataset Used,
Evaluation Metrics, Model Accuracy, Learning Task, and Application Field.
Model Year Input Data Scalability, Inter-
pretability, Compu-
tational Efficiency,
ReproducibilityDataset Evaluation
MetricsModel Accuracy Application
FFM[354] 2016 Categorical,
Numeric, Single
Field-, Low, High, High Criteo, Avazu Logloss, Rank Criteo: 0.44603/3
Avazu: 0.38205/3Advertising
NFM[98] 2017 Context features
in one hot encod-
ing-, Low, High, High Frappe, MovieLens RMSE Frappe: 0.3095
MovieLens: 0.4443e-commerce
LTMF[355] 2018 User-item inter-
action-, Medium, -, Low 8 Amazon subsets
(AFA, BB, MI, OP,
PS, VG, PLG, DM,
AIV , GGF)MSE DM: 0.965
AIV: 1.309
GGF: 1.386e-commerce
DeepFM[99] 2017 Implicit interac-
tion-, Low, Low, High Criteo, Company AUC, LogLoss Criteo:
0.8007/0.45083
Company:
0.8715/0.02618Web applica-
tions
xDeepFM[ 141]2018 Implicit, explicit
interactions-, -, Low, High Criteo, Dianping,
Bing NewsAUC, Logloss Criteo:
0.8012/0.4493
Dianping:
0.8576/0.3225
Bing News:
0.8377/0.2662e-commerce
AFM[356] 2017 Implicit interac-
tion-, High, -, High Frappe, MovieLens RMSE Frappe: 0.3102
MovieLens: 0.4325e-commerce,
Online adver-
tising, Image
recognition
GCM[188] 2022 User-item graph -, Medium, Medium,
HighYelp-NC, Yelp-OH,
Amazon-bookHr@50,
NDGC@50Yelp-NC:
0.2421/0.0854
Yelp-OH:
0.5166/0.2008
Amazon-book:
0.2232/0.0810e-commerce
ACA-
GRU[226]2020 Implicit interac-
tion-, High, -, Low MovieLens-100K,
MovieLens-1K,
NetflixR@10, P@10,
F1@10, MAPMovieLens-1M:
0.2207/0.0630/
0.0980/0.2432
Netflix:
0.2308/0.0659/
0.1025/0.2620e-commerce
PreADBC
ACF[225]2020 implicit interac-
tion, context-, -, -, Low YFCC100M AP, MAP, Re-
call, F1, nDGC0.3542/0.3903/
0.8292/0.4/ 0.6741e-tourism
BERT-
GCN[134]2020 node graph &
node-node inter-
actionScale, High, -, High AAN, FullTextPeer-
ReadMAP, MRR,
Recall@80AAN:
0.6189/0.6036/
0.8538
FullTextPeerRead:
0.4181/0.4179/
0.6994Paper recom-
mendation
SCR[237] 2024 User prefer-
ences-, High, Low, Low Gowalla, BrightKite HR, MRR Gowalla:
0.4804/0.2143
BrightKite:
0.5721/0.2856Location-
based social
networks
8.2 Review-based Recommender Systems
A review-based RS uses textual reviews and ratings from users to generate personalized recommendations for products
or services [ 358,45]. The review-based RS have evolved by improving through various models. Initially, models like
Hidden Factors as Topics (HFT) [ 359] aligned topics from reviews with latent dimensions from ratings. Successive
approaches, such as Rating-Boosted Latent Topics (RBLT) [ 360], Topic Initialized Latent Factor Model (TIM) [ 307],
and deep learning models like Convolutional Matrix Factorization (ConvMF) [ 361] and Deep Cooperative Neural
Networks (DeepCoNN) [ 87], utilized neural networks to better handle sparse data and extract nuanced features from
reviews. Advanced models, including SentiRec [ 362] and Neural Collaborative Topic Regression (NCTR) [ 363],
incorporated sentiment analysis and hybrid data integration to refine recommendations further.
28
Page 29:
arXiv Template A P REPRINT
Attention-based models, such as Adaptive Aspect Attention Model (A3NCF) [ 364], Attentive Aspect Modeling for
Review-aware Recommendation (AARM) [ 365], and Cross-Modality Mutual Attention (NRCMA) [ 366], have used
aspect-specific attention to prioritize relevant features, enhancing both precision and personalization of recommendations.
Techniques like Neural Networks with Dual Local and Global Attention (D-Attn) [ 367], Neural Attentional Rating
Regression (NARRE) [ 368], and Neural Recommendation with Personalized Attention (NRPA) [ 369] have focused on
integrating personal attention and dual learning mechanisms to improve recommendation accuracy.
Topical Attention Regularized Matrix Factorization (TARMF) [ 370], Asymmetrical Hierarchical Networks (AHN)
[371], and Reliable recommendation with review-level (RRRE) [ 129] have integrated user reviews with advanced neural
and attention mechanisms to further boost RS efficiency. Emerging graph-based methods like Heterogeneous Graph
Neural Recommender (HGNR) [ 172], Aspect-Aware Higher-Order Representations (AHOR) [ 372], and Multi-aspect
Graph Contrastive Learning (MAGCL) [ 205] have tackled data sparsity and semantic complexity by employing GNNs
to enhance the overall recommendation framework.
Practical Challenges Addressed Review-based RS have impacted various industries by leveraging user-generated
content to enhance the personalization and relevance of recommendations. Industries ranging from e-commerce and
hospitality to digital media and services benefit from these systems by providing more targeted offerings, which can
lead to increased sales and customer satisfaction. Additionally, by interpreting complex user feedback, these systems
contribute to product development and refinement, helping businesses better understand market demands and customer
concerns. More details in Table 13.
Table 13: Comprehensive Overview of Review Based Models across Various Metrics and Use Cases. This table details
each model’s Input features, Year of Publication, and Characteristics such as Scalability, Interpretability, Efficiency, and
Reproducibility (rated as High, Medium, or Low). It also lists the Dataset Used, Evaluation Metrics, Model Accuracy,
Learning Task, and Application Field.
Model Year Input Data Scalability, Inter-
pretability, Compu-
tational Efficiency,
ReproducibilityDataset Evaluation
MetricsModel Accuracy Application
MAGCL[ 205]2018 features -,-,-, Low Amazon (Music,
Toy, CD), YelpMRR, nDCG Music:
0.2841/0.3562,
Toy: 0.1802/ 0.2281,
CD: 0.4110/0.4863,
Yelp: 0.2899/
0.3597e-commerce
HFRT [359] 2013 features -, -, -, High51Amazon (total);
Pubs (Ratebeer);
Beer (Ratebeer);
Pubs (Beeradvo-
cate); Beer (Beer-
advocate); Wine
(Cellartracker);
Citysearch; Yelp
PhoenixMSE Amazon (total):
1.3290; Pubs (Rate-
beer): 0.4560; Beer
(Ratebeer): 0.3010;
Pubs (Beeradvo-
cate): 0.3110; Beer
(Beeradvocate):
0.3670; Wine (Cel-
lartracker): 0.0280;
Citysearch: 1.7280;
Yelp Phoenix:
1.2250E-commerce,
Review
Platforms
RBLT [360] 2016 features -, -, -, - 26 Amazon datasets MSE N/A E-commerce
TIM [360] 2020 features -, -, -, - Amazon Toys &
Games; Amazon Pet
Supplies; Amazon
Health & Personal
Care; TripAdvisor
HotelsRecall; Hit
Ratio; NDCG;
PrecisionAmazon Toys &
Games: 0.264,
0.535, 0.169,
0.076; Amazon Pet
Supplies: 0.362,
0.660, 0.215, 0.097;
Amazon Health
& Personal Care:
0.316, 0.614, 0.190,
0.085; TripAdvisor
Hotels: 0.581, 0.702,
0.260, 0.078E-commerce;
Hospitality
ConvMF
[361]2016 features -, -, -, High52MovieLens; Ama-
zon Instant VideoRMSE MovieLens: 0.8531;
MovieLens 10m:
0.7958; Amazon
Instant Video:
1.1337E-commerce;
Movies
51https://github.com/mcauley-sd/HFRT
52http://dm.postech.ac.kr/ConvMF
29
Page 30:
arXiv Template A P REPRINT
Table 13 – continued from previous page
Model Year Input Data Scalability, Inter-
pretability, Compu-
tational Efficiency,
ReproducibilityDataset Evaluation
MetricsModel Accuracy Application
DeepCoNN
[87]2017 features -, -, -, Low Yelp; Amazon; Beer MSE Yelp: 1.441; Ama-
zon: 1.268; Beer:
0.273Diverse
(Restaurants;
General
Products;
Beverages)
TransNets
[373]2017 features -, -, -, Low Yelp17; AZ-Elec;
AZ-CSJ; AZ-MovMSE Yelp17: 1.5913; AZ-
Elec: 1.7781; AZ-
CSJ: 1.4487; AZ-
Mov: 1.2691E-commerce
SentiRec
[362]2020 features -, -, -, - MSN News MSE; AUC;
MRR;
nDCG@5;
nDCG@10MSE: -; AUC:
0.6294; MRR:
0.3013; nDCG@5:
0.3237; nDCG@10:
0.4165Online News
Services
A3NCF
[364]2018 features -, -, -, Low Baby; Grocery;
Home & Kitchen;
Garden; Sports;
Yelp2017RMSE Baby: 1.082; Gro-
cery: 0.985; Home
& Kitchen: 1.051;
Garden: 1.021;
Sports: 0.940;
Yelp2017: 1.152E-commerce;
Local Busi-
ness Reviews
AARM [ 365] 2019 features -, -, -, Low Movies and TV;
CDs and Vinyl;
Clothing, Shoes
and Jewelry; Cell
Phones and Acces-
sories; BeautyNDCG; HT;
Recall; Preci-
sionMovies and TV:
5.020, 15.187,
7.140, 1.834; CDs
and Vinyl: 7.252,
20.749, 9.965,
2.716; Clothing,
Shoes and Jewelry:
1.957, 4.915, 3.292,
0.511; Cell Phones
and Accessories:
4.976, 11.568,
8.014, 1.259;
Beauty: 5.314,
13.648, 7.947, 1.818E-commerce
D-Attn [367] 2017 features -, -, -, Low Yelp; Amazon MSE Yelp: 1.191; Ama-
zon: 0.855E-commerce
NARRE
[368]2018 features -, -, -, Low Amazon Toys &
Games, Kindle
Store, Movies & TV;
Yelp 2017RMSE Toys & Games:
0.8769; Kindle
Store: 0.7783;
Movies & TV:
0.9965; Yelp 2017:
1.1559E-commerce;
Restaurant
Reviews
NRPA [369] 2019 features -, -, -, Low Yelp 2013, Yelp
2014, Amazon
Electronics, Ama-
zon Video Games,
Amazon Gourmet
FoodsMSE Yelp 2013: 0.872;
Yelp 2014: 0.897;
Amazon Electronics:
1.047; Amazon
Video Games:
1.014; Amazon
Gourmet Foods:
0.953Service
Reviews;
Consumer
Electronics;
Video Games;
Gourmet
Foods
DAML [ 374] 2019 features -, -, -, Low Musical Instru-
ments, Office
Products, Grocery
and Gourmet Food,
Video Games,
Sports and OutdoorsMAE Musical Instru-
ments: 0.6510;
Office Products:
0.6124; Grocery
and Gourmet Food:
0.7354; Video
Games: 0.7881;
Sports and Out-
doors: 0.6676E-commerce
MrRec [375] 2020 features -, -, -, Low Amazon Books,
Digital Ebook Pur-
chase, Digital Music
Purchase, Digital
Video Download,
Mobile Apps, Mu-
sic, Toys, Video
DVD; GoodreadsMSE Books: 1.307; Digi-
tal Ebook Purchase:
1.253; Digital Mu-
sic Purchase: 1.682;
Digital Video Down-
load: 1.288; Mobile
Apps: 1.036; Music:
1.269; Toys: 1.392;
Video DVD: 1.243;
Goodreads: 1.189Multilingual
E-commerce
30
Page 31:
arXiv Template A P REPRINT
8.3 Aspect-based Recommender Systems
Aspect-based RS extract and analyze specific product attributes from reviews, providing tailored recommendations to
the users based on item aspects [ 376]. This approach to RS differs with review-based RS, which assess overall user
sentiment and preferences from review content.
Aspect-based RS have evolved from traditional RS that rely on user-item interactions to methods that delve into item
aspects or features for tailored suggestions. Early works laid the foundation by extracting aspect-related information
from reviews to enhance user satisfaction and uncover niche items [ 377]. The concept of multi-criteria RS that uses
CF and opinion mining to extract aspects and sentiment from user reviews, shows better accuracy over single-criteria
methods [ 378]. The Aspect-based Neural Recommender (ANR) uses representation learning for users and items [ 376].
Simultaneously, introduction of lightweight ontologies in aspect-based RS improve the search for relevant venues [ 379].
The integration of deep learning methods in the aspect-based RS, such as in related works [ 380,381,382], enable the
capturing of syntactic and semantic features without extensive feature engineering in these methods [56].
Aspect-based sentiment analysis began to play a critical role in detecting sentiment polarity towards specific aspects
within a context, exemplified by Sentic GCN [ 175] and sentiment-analysis with CF [ 382]. Lately, incorporating
neural co-attention mechanisms and deep neural networks further refine the consideration of user aspects in making
recommendations [ 383]. Multi-criteria RS such as Hybrid Aspect-based Latent Factor Model (HALFM) [ 384] and
Aspect-based Opinion mining using Deep learning method for RS ( AODR ) [ 380], which utilized global and local
aspect-based latent factor models and weighted aspect-based opinion mining further improve recommendation accuracy.
Specialized approaches like the use of a query-click bipartite graph alongside an iterative clustering algorithm start
recommending products for specific events and focus on event-related aspects [ 118]. The integration of diversity
preference in link recommendations for online social networks highlight the ongoing evolution and expansion of
aspect-based RS [385].
Aspect-based RS has applications mainly in tourism [ 386] and customer-generated content, such as for restaurants
[387].
Practical Challenges Addressed Aspect-based RS effectively address several practical challenges by focusing on
specific product attributes extracted from user reviews. These systems enhance personalization by tailoring recommen-
dations based on individual user preferences and item characteristics. In e-commerce, aspect-based RS can recommend
niche products by analyzing detailed aspects like product features and user sentiments. This capability improves
customer satisfaction and boosts sales by aligning recommendations more closely with user needs. Additionally, in
the tourism and hospitality industries, aspect-based RS provide recommendations by considering specific attributes of
destinations or services, thus offering more relevant and satisfactory suggestions. More details in Table 14.
Table 14: Comprehensive Overview of Aspect Based Models across Various Metrics and Use Cases. This table details
each model’s Input features, Year of Publication, and Characteristics such as Scalability, Interpretability, Efficiency,
and Reproducibility (rated as High, Medium, or Low). It also lists the Dataset Used, Evaluation Metrics, Model
Accuracy, Learning Task, and Application Field. Metrics that their numerical value is not reported are specified with
“No numerical value”.
Model Year Input Data Scalability, Inter-
pretability, Efficiency,
ReproducibilityDataset Evaluation
MetricsModel Accuracy Application
ANR [376] 2018 user-item inter-
actionHigh, -,-, High Amazon, Yelp MSE No numerical value e-commerce
SULM [377] 2017 Sentiment analy-
sis-, -, Medium, No Yelp: restaurants,
hotels, beauty & spaPrecision@Top3,
AUCRestaurants: 0.8180,
0.7070
Hotels: 0.8490,
0.7450
Beauty & spa:
0.8620, 0.6630E-commerce
[378] 2017 Multi-criteria
CF; aspect-
based sentiment
analysisMedium, Medium,
Medium, NoYelp; TripAdvisor;
AmazonMAE Yelp: 0.8362
TripAdvisor: 0.7111
Amazon: 0.6276E-commerce
[379] 2018 Aspect extrac-
tion; content-
based filtering-, -, Medium, No Restaurant/museums
reviewsF1 score 0.7026
Museums: N/aTourism
AODR [380] 2020 Opinion mining High, -,-, High Amazon, Yelp RMSE, MAE,
Prec@10,
MAPNo numerical value E-commerce
31
Page 32:
arXiv Template A P REPRINT
Table 14 – continued from previous page
Model Year Input Data Scalability, Inter-
pretability, Efficiency,
ReproducibilityDataset Evaluation
MetricsModel Accuracy Application
REAO [381] 2020 Aspect-based
opinion mining;
deep learningHigh, Medium, High,
NoSemEval2014
Restaurant; Se-
mEval2014 Laptop;
Amazon Musical In-
struments; Amazon
Automotive; Ama-
zon Pet Supplies;
Amazon Video
Games; Amazon
Instant Video; YelpRMSE; MAE MI: 0.8020, 0.6320
Auto: 0.8140,
0.5980
IV: 0.9740, 0.7840
Pet: 0.9720, 0.7840
V .Games: 1.0270,
0.8170
Yelp: 1.1310,
0.9410E-commerce
SE-DCF
[382]2021 Sentiment En-
hanced Deep
Collaborative
FilteringMedium, Medium, High,
NoAmazon fine food;
Amazon toys and
games; Amazon
clothing, shoes and
jewelleryMAE; RMSE Amazon fine food:
0.1562, 0.2771
Amazon toys and
games: 0.1625,
0.2819
Amazon clothing,
shoes and jewellery:
0.1528, 0.2772E-commerce
Sentic GCN
[175]2022 Graph convolu-
tionalHigh, Mid,High, High SemEval Accuracy,
Macro-F1No numerical value General
ANR-AP
[383]2023 Neural Recom-
mender; Adap-
tive PredictionMedium, Medium, High,
NoAmazon movie
dataset (1996-2014);
Amazon dataset
(web-scraped)Precision@k;
Recall@k;
F1@kTop 5: 0.4421,
0.1790, 0.2517
Top 10: 0.4420,
0.3580, 0.3897
Top 20: 0.3230,
0.4421, 0.3674E-commerce
HALFM
[384]2020 Hybrid High, Mid,High, High Amazon MSE Outperforms most Personalized
Event-based
PCR [118]2021 Click Graph-
based Cluster-
ingHigh, Mid,High, High Walmart Precision, Het-
erogeneity, Co-
hesionHigh precision, ef-
fective aspect clus-
teringE-commerce
DPA-LR
[385]2023 Diversity
preference-
aware link
recommenda-
tionMedium, Medium, High,
NoGoogle+; Major U.S.
social networkDPMS; Preci-
sion; Recall; F1
ScoreGoogle+: 0.4559,
0.1541, 0.1559,
0.1149Social net-
works
Emotion-
ABSA [386]2023 Emotion and
sentimentHigh, -,-, High User-generated Emotion analy-
sisimprovement Tourism
ABSA-CSF
[387]2023 Sentiment analy-
sis; Conditional
Survival ForestMedium, Medium, High,
NoYelp C-index; IBS Yelp: 0.7370,
0.0387Tourism
8.4 Explainable and Trustworthy Recommender Systems
To gain user engagement and satisfaction, latest works in RS start prioritizing transparency and trustworthiness. An
explainable RS provides transparent recommendations by offering clear, understandable reasons behind its suggestions,
enhancing user trust and system usability [ 31]. In parallel, a trustworthy RS reliably produces accurate and fair
recommendations to ensure ethical practices like privacy protection and bias minimization to maintain user confidence
[388].
Advancements in explainable and trustworthy RS have evolved, starting with phrase-level analysis of user reviews to
enhance recommendation explainability by identifying critical item aspects [ 389]. Subsequent models like Tripartite
Graph Ranking (TriRank) have improved top-N recommendations by extracting aspects from reviews and creating a
user-item-aspect ternary relation [ 390]. Concurrently, models such as the Tree-Enhanced Embedding Model (TEM)
merge embedding-based and tree-based methods with an attention network to ensure transparency, utilizing rich side
information and explicit decision rules [ 391]. This integration extends to combining CF with structured knowledge bases
and unstructured data like textual reviews for personalized and understandable recommendations [ 392]. Additionally,
techniques like RL have been applied to generate flexible, high-quality explanations across recommendation models
[251].
Further developments include the Multi-Modal Aspect-aware Topic Model (MATM), which utilizes multi-modal data
for detailed explanations reflecting diverse user preferences [ 393]. A variety of approaches, including natural language
models, counterfactual reasoning, and visual explanations, have been employed to enhance interaction, fairness, and
personalization in RS [394, 299, 183, 395, 396].
Recent efforts like the Counterfactual Explainable Fairness (CEF) framework focus on identifying and mitigating
fairness issues in RS [ 397]. Discussions around Trustworthy RS further emphasize the critical dimensions of Safety
32
Page 33:
arXiv Template A P REPRINT
& Robustness, Fairness, Explainability, Privacy, and Accountability, vital for maintaining the integrity and reliability
of RS [ 388]. These developments show the growing importance of creating RS that are not only effective but also
equitable and trustworthy.
Practical Challenges Addressed Explainable and trustworthy RS enhance industry practices by providing transparent
and personalized recommendations based on user reviews and sophisticated models. These systems increase customer
trust and satisfaction by explaining recommendation logic, which is very important in industries like e-commerce,
tourism, and hospitality. These systems can be used along with regular RS processes for better customer experiences.
Table 15: Comprehensive Overview of Explainable and Trustworthy Recommender System Models across Various
Metrics and Use Cases. This table details each model’s Input features, Year of Publication, and Characteristics such as
Scalability, Interpretability, Efficiency, and Reproducibility (rated as High, Medium, or Low). It also lists the Dataset
Used, Evaluation Metrics, Model Accuracy, Learning Task, and Application Field.
Model Year Input Data Scalability, Inter-
pretability, Efficiency,
ReproducibilityDataset Evaluation
MetricsModel Accuracy Application
EFM[389] 2014 user-item inter-
actionMid, High, Medium,
HighYelp, Dianping RMSE,
NDCG@501.212, 0.284;
0.9222, 0.284e-commerce
TriRank[ 390] 2015 user-item-
aspect interac-
tionHigh, High, High, High Yelp, Amazon Elec-
tronicsHR@50,
NDCG@5018.58,7.69;
18.44,12.36e-commerce
TEM[391] 2018 user-item inter-
actionHigh, High, High, High LON-A, NYC-R Logloss,
NDCG@50.0791,0.1192;
0.6828,0.4038Tourism,
restaurant
ECFKG
[392]2018 knowledge
graph embed-
dingsHigh, High, Medium,
HighAmazon (Clothing,
Beauty)NDCG, Recall,
Prec.3.091,5.466,0.763;
6.399,10.411,1.986e-commerce
MMALFM
[393]2019 user-item inter-
actionHigh, High, Medium,
HighYelp, Amazon NDCG, Preci-
sionMultiple e-commerce,
restaurant
PGPR [213] 2019 kg-based path
reasoningHigh, High, Medium,
HighAmazon (various do-
mains)NDCG, Recall,
HR, Precisiongenerally high per-
formancee-commerce
PETER
[395]2021 user-item inter-
actionHigh, High, Medium,
HighYelp, Amazon, Tri-
pAdvisorRMSE, MSE 1.01,0.95,0.81;
0.78,0.71,0.63e-commerce,
restaurant
CEF [397] 2022 user-item inter-
action-, High, - , High Yelp, Amazon Precision, Re-
call, F1 ScoreMultiple e-commerce,
restaurant
PEPLER
[398]2023 user-item inter-
actionHigh, High, High, High Yelp, Amazon, Tri-
pAdvisorBLEU,
ROUGE,
USRoutperforms base-
linese-commerce,
restaurant
ExpGCN
[399]2023 user-item inter-
actionHigh, High, High, High Yelp, Amazon, Tri-
pAdvisor, HotelRecRecall, NDCG outperforms base-
linese-commerce,
restaurant
8.5 Fairness, Accountability, Transparency, and Ethics (FATE) in Recommender Systems
There is a growing focus on Fairness, Accountability, Transparency, and Ethics (FATE) in RS, which ensures that RS
are fair to all users, responsible for their recommendations, transparent in how decisions are made, and ethically aligned
with institutional or societal values [400].
Fairness in RS, as outlined in [ 401], refers to the ethical principle and requirement that recommender algorithms allocate
resource (information, opportunities, or exposure) in a manner that is equitable and just across different users and items.
The evolution of fairness methods in RS shows a shift from simple pre-processing strategies to in-processing (model
adjustments) and post-processing techniques.
Pre-processing Fairness Methods Pre-processing efforts for fairness in RS involve adjusting training data, altering
proportions of protected groups (like gender, race, age) through resampling [ 174] or adding synthetic data [ 402]. These
methods aim to mitigate biases in input data before model training, they struggle to entirely eliminate biases that appear
during training or inference.
In-processing Fairness Methods In-processing fairness methods in RS primarily utilize ranking approaches and advanced
techniques to incorporate fairness directly into model training, yielding more immediate improvements by modifying
elements closely tied to the final output. Regularization techniques play a crucial role by embedding fairness constraints
or penalties into the objective function to balance accuracy with fairness, with strategies ranging from employing
fairness metrics as regularization [ 403], using distribution matching [ 404], enforcing orthogonality between insensitive
and sensitive factors [ 405], to adding pairwise fairness regularization [ 406] and applying F-statistic of ANOV A [ 407],
along with integrating normalization terms [408, 409].
Adversarial learning further enhances fairness by learning representations that maintain independence from sensitive
attributes or ensure equitable distribution across groups, with notable applications in graph embeddings [ 178], score
33
Page 34:
arXiv Template A P REPRINT
distribution similarity enhancement [ 410], graph-based recommendations [ 200], and personalized counterfactual
fairness [ 411]. Reinforcement learning approaches [ 412] introduce fairness through rewards and constraints, aiming
for sustainable fairness. Additional in-processing methods include adding noise to Variational Autoencoders (V AEs)
[413], utilizing pre-training and fine-tuning with bias correction techniques [ 414], and adjusting gradients for fair
distribution [ 415]. In-processing methods enhance fairness directly but may face performance degradation due to
additional constraints and can be affected by subsequent re-ranking stages, altering intended outcomes.
Post-Processing Fairness Methods Post-processing methods involve adjusting the initial output of a recommendation
model to satisfy certain fairness criteria before presenting the final recommendations to users. These methods typically
act as a post-processing step, optimizing the balance between recommendation relevance and fairness after the primary
ranking algorithm has made its predictions. Slot-wise re-ranking methods aim to balance ranking utility with fairness
constraints across various contexts. These methods include employing two queues for group fairness [ 416] and calibrated
recommendations [ 417], enhancing group fairness through interval-constrained sorting [ 418], personalized fairness-
aware re-ranking [ 419]. User-wise re-ranking approaches, on the other hand, consider individual user perspectives
[420]. Global-wise re-ranking strategies seek broader fairness solutions, adopting methods for equitable explainability
and maximum flow principles [ 171]. These global approaches ensure fairness not just for current users and providers,
but also aim to include fairness among new items [ 421]. Recent surveys [ 401,422,423,224] have emphasized the
growing importance of fairness in RS .
Practical Challenges Addressed In the e-commerce industry, FATE-based RS contribute to building customer trust
and enhance the shopping experience. These systems are designed to mitigate biases and ensure fairness in product
recommendations, which helps retain a diverse customer base and comply with increasing regulatory requirements
for ethical AI practices. By integrating FATE principles, these RS not only boost customer satisfaction but also foster
a responsible brand image, which is essential for long-term business success. FATE-based RS can be seamlessly
used alongside regular RS processes to enhance transparency and accountability, thereby improving overall customer
engagement and loyalty. More details are provided in Table 16.
Table 16: Comprehensive Overview Recommender System Models for FATE across Various Metrics and Use Cases.
This table details each model’s Input features, Year of Publication, and Characteristics such as Scalability, Interpretability,
Efficiency, and Reproducibility (rated as High, Medium, or Low). It also lists the Dataset Used, Evaluation Metrics,
Model Accuracy, Learning Task, and Application Field.
Model Year Input Data Scalability, Inter-
pretability, Efficiency,
ReproducibilityDataset Evaluation
MetricsModel Accuracy Application
Antidote
Data
Adding[402]2019 user-item inter-
action- , -, High, Low MovieLens Polarization,
unfairnessNone e-commerce
Beyond Par-
ity [403]2017 user-item inter-
action-, -, High, Low MovieLens Error, unfair-
ness0.887, 0.010 e-commerce
IERS[404] 2018 user-item inter-
action-, -, High, Low MovieLens, Flixter,
SushiMAE, degree
of indepen-
denceMovielens: 0.7/
0.01, Flixter: 0.65/
0.01, Sushi: 0.92/
0.05e-commerce
Fairness-
aware
TR[405]2018 user-item inter-
action- , - , -, Low MovieLens, Twitter Precision@15,
Recall@15Movielens: 0.032/
0.08, Twitter:
0.03298, 0.0687e-commerce,
Social Net-
works
Fairness
Pairwise
Comparisons[ 406]2019 user-item inter-
action-, - , -, Low Synthetic data Overall Pair-
wise accuracy,
intra-group
Pairwise
Accuracy35.6%, 16.7% e-commerce
MarketBias[ 407]2020 user-item inter-
action-, - , -, Low ModCloth, Electron-
icsMSE, MAE ModCloth: 1.176/
0.859, Electronics:
1.590/ 1.025e-commerce
Latent factor
model [408]2020 user-item inter-
actionHigh, - , -, High New York Times F1@10,
F1@20,
F1@50,
F1@1000.5458, 0.5425,
0.5405, 0.5401News recom-
mendation
News Bias
Reduction
[409]2023 user-item inter-
actionHigh, -, -, High MIND-small, Out-
brainPrecision@5,
Recall@5,
NDCG@5,
Gini IndexMIND-small: 0.65/
0.55/ 0.60/ 0.18,
Outbrain: 0.62/
0.52/ 0.57/ 0.19News recom-
mendation
Fairness-in-
Cold-Start
[421]2023 user-item inter-
action-, - , -, High Movielens1M,
Movielens20M,
CiteULike, XINGNDCG@15,
NDCG@30Movielens1M:
0.5516/ 0.5332,
Movielens20M:
0.4408/ 0.4308,
CiteULike: 0.2268/
0.2670, XING:
0.2251/ 0.2762News recom-
mendation
34
Page 35:
arXiv Template A P REPRINT
Table 16 – continued from previous page
Model Year RS Type Scalability, Inter-
pretability, Efficiency,
ReproducibilityDataset Evaluation
MetricsModel Accuracy Application
FCPO [412] 2021 user-item inter-
action-, - , -, High Movielens100k,
Movielens1MRecall@5,
F1@5,
NDCG@5,
Gini Index@5Movielens100k:
4.740/ 4.547/ 6.031/
98.73, Movie-
lens1M: 2.033/
2.668/ 4.398/ 99.81e-commerce
Long Term
Fairness
[413]2019 user-item inter-
action-, - , -, High Movielens, Netflix,
MSDNDCG@100 Movielens: 0.999,
Netflix:0.999,
MSD:0.998e-commerce
NFCF [414] 2021 user-item inter-
action-, - , -, High Movielens, Face-
bookMovielens:
HR@5,
NDCG@5,
Facebook:
HR@10,
NDCG@10Movielens: 0.670,
0.480, Face-
book:0.551, 0.326e-commerce
Contextualized
Fairness
[415]2022 user-item inter-
action-, - , -, Low XING HR@5,
NDCG@50.581, 0.47 e-commerce
FAIR[416] 2017 user-item inter-
action-, - , -, Low COMPAS, Ger.
credit, SAT, XINGNDCG 0.9858, 0.9983,
0.9996, 1.0000e-commerce
LinkedIn Tal-
ent Solutions
[132]2018 user-item inter-
action-, - , -, Low - - - e-commerce
8.6 Miscellaneous
There are also other RS that can serve specialized purposes, as outlined briefly below.
Group-based RS are designed to provide collective recommendations by considering users’ shared preferences, social
dynamics, and behavioral aspects [ 424]. Initial studies address the cold-start problem with group-specific methods and
deep learning applications [ 425]. Subsequent research emphasizes the importance of diversity, introducing algorithms
to optimize group utility and variety [ 426]. Advancements in group recommendations explore trust and social dynamics
by using social influence and preference relation-based frameworks [427, 428, 429].
Some work aggregates user preferences into a unified group preference, using both explicit and implicit feedback
mechanisms [ 424]. Additionally, context-aware capabilities considering significant factors for group-based scenarios
are highlighted [ 133]. Strategies for aggregating individual preferences, such as aggregated voting and ensuring
satisfaction for all members, are also addressed [ 430,431]. Recent research presents novel approaches to maximize
group satisfaction through least misery methods, reflecting ongoing refinement to better cater to group needs [104].
There are also other methods, such as the Multi-Stakeholder RS approach [ 432] that acknowledges that recommendations
often affect multiple stakeholders beyond the immediate users. For example, in a movie recommendation scenario,
stakeholders include not only the viewers but also the content creators, distributors, and platforms hosting the content.
Social RS [ 433,434] target the social media domain to cope with the social overload challenge by presenting the most
relevant and attractive data to users, typically through the application of personalization techniques. Interactive and
Conversational RS [ 40] engage users (or groups of users) in a dialogue to iteratively refine recommendations based
on feedback. This approach is particularly useful in group settings, where initial recommendations may need to be
negotiated among members through a series of interactions.
Overall, these methods in group-based and social RS reflect a commitment to improving both the precision and
satisfaction of group recommendations in increasingly complex scenarios.
9 Applications of Recommender Systems Across Different Domains
This section explores the technological developments and specific applications of RS in various domains. The goal is to
highlight how advancements in areas such as GNNs, RL, LLMs, multimodal and related methods are being applied to
tackle domain-specific challenges.
E-commerce/E-Business In the digital era, e-commerce platforms utilize RS to personalize the shopping experience
by recommending products based on individual preferences, browsing and purchase histories, and cart contents, thus
enhancing user engagement and driving sales growth [ 105]. Advances include the integration of big data and ML to
improve satisfaction on platforms like Amazon [ 435], and Alibaba [ 227]. Techniques such as CF and CBF, along with
35
Page 36:
arXiv Template A P REPRINT
newer methods like graph-based models and hypergraph ranking, refine user preference predictions [ 303,207,211].
Sophisticated technologies like deep learning, deep reinforcement learning, and GNNs now capture complex user
behaviors [ 103,181,127,43]. Despite these advancements, challenges like information overload and the focus on
click-through rates persist, necessitating smarter, multi-objective RS approaches [436, 437].
E-Entertainment (Music, Movies, Games, Dating Apps) Platforms like Netflix and Spotify personalize content
recommendations using a mix of CF, CBF, and hybrid approaches, employing deep learning and ML to tailor suggestions
based on user interactions and contextual factors [ 438,439]. Netflix utilizes deep learning and a blend of CF and CBF
to analyze users’ interactions and viewing habits [ 439,73], while Spotify leverages ML and NLP, introducing systems
like GNN for audiobooks to address data sparsity and enhance content discovery [ 440,209,441]. The video game
industry, exemplified by STEAM, uses advanced models to offer personalized game suggestions [ 442], addressing
broader implications through multi-stakeholder recommendations [ 432]. RS also leverage multimedia content for
diverse recommendations [ 20]. Innovations such as GNNs and knowledge-based methods improve personalization
[440], but challenges in dynamic consumer preferences and the need for explainability in RS remain [41].
E-Health Health RS analyze health data, lifestyle, and genetics to enhance outcomes [ 443]. They address challenges
like privacy and trust, and are integral to Healthcare 4.0, focusing on personalized interventions [24, 81, 22, 444].
Systematic reviews assess health RS progress and emphasize risk management and privacy [ 7,445,446,447,26].
Advances in ML and deep learning have improved RS, with applications in diabetes, cardiac care, and beyond
[448, 449, 450, 110].
Advacements in algorithms include enhanced prediction accuracy through trust relationships and advanced ML
techniques such as hybrid deep learning models [ 451,452,453,454,455,456,457,458]. Emerging research explores
continual learning and clustering-based techniques for improved clinical RS applications [459, 460, 461].
E-Government RS E-government utilizes electronic communication technologies to enhance service delivery, citizen
engagement, and internal processes, integrating RS to improve user experience through AI and machine learning
[115,462,25]. These systems play a crucial role in smart cities by supporting information filtering, stakeholder
engagement, and decision-making [128].
Initial development of RS in e-government used CF and CBF, incorporating hybrid models for more accurate predictions
[463,464,465,466]. The use of NLP and predictive analytics enhances public service recommendations [ 467,107].
Challenges such as information overload are addressed by improving CF with negative item techniques, while newer
methods like CNNs and GNNs advance feature extraction and recommendation accuracy in industrial applications
[465, 468, 185].
E-Library and E-Learning E-learning, a subset of e-libraries, utilizes electronic resources (e-books, academic papers,
journals, and other digital content) for learning and includes a broader range of digital services for information retrieval
and research [ 116]. Early development in e-library RS focused on hybrid systems combining CBF and CF techniques,
often featuring bookshelf functionalities to personalize interactions [ 469,470]. These systems also use bibliographic
network representation models for citation recommendations [ 177,471,64,472]. Advances in deep learning and
context-aware recommendations have significantly improved the efficiency of e-learning systems, surpassing traditional
methods [473, 82].
E-Tourism/Travel RS have transformed travel and tourism by using vast data to provide personalized travel sugges-
tions, thus enhancing user satisfaction [ 80,21,474]. Major platforms like TripAdvisor and Booking.com employ CF,
CBF, and hybrid methods to offer tailored travel options [ 475,476]. Continuous advancements are needed to manage
dynamic data and maintain up-to-date, transparent recommendations that build user trust [ 62,477]. Future innovations
may incorporate immersive destination previews, further personalizing travel experiences [210, 478].
E-Finance RS in finance assist investors by aligning investment options with individual goals and risk tolerance,
significantly enhancing investor engagement and informed decision-making through analysis of financial history,
risk profiles, market trends, and economic indicators [ 19,479]. Notable implementations like the FinPathlight [ 480]
framework enhance financial literacy and capability, while integrating behavioral finance [ 481] integrates behavioral
finance to tailor financial advice based on psychological biases. Additionally, platforms like StockTwits use sentiment
analysis for more accurate investment recommendations [ 479], and KiRTi employs blockchain and deep learning to
automate and secure lending processes [ 482]. These technologies collectively improve the personalization of financial
services, advice, and strategy optimization [123].
36
Page 37:
arXiv Template A P REPRINT
Despite progress, challenges remain in handling market volatility and ensuring transparency and trust in RS [ 19]. Future
developments may focus on enhancing explainability and employing predictive analytics to better anticipate market
trends and user preferences, further personalizing financial advice [483].
E-News News RS curate and suggest content to users based on methodologies like CF, CBF, and hybrid approaches,
distinguishing between personalized and non-personalized systems [ 59,60]. Significant advancements in news RS
have integrated deep learning and ML to improve how news content and user data are modeled. This includes using
neural network architectures and pre-trained language models to enhance the accuracy of content recommendations
[152,484,485]. New techniques also explore the use of GNNs to understand complex user-news interactions [ 220,192]
and innovative models like Prompt4NR for advanced click prediction tasks [ 346]. The development of news RS also
faces ethical challenges, such as addressing filter bubbles, ensuring diversity, and promoting fairness, which are crucial
for maintaining user trust and system integrity [486, 408, 487].
Miscellaneous Numerous platforms have leveraged advanced RS technologies to enhance user engagement and
content personalization. YouTube employs deep neural networks to refine its recommendation process, focusing on
optimal ranking and selection of videos [ 488]. Google Play utilizes both linear models and neural networks within
its Wide & Deep Learning framework to achieve a balance between memorization and generalization [ 97]. LinkedIn
enhances job and content recommendation using real-time processing and scoring mechanisms, integrating CF and deep
learning to match job seekers with suitable opportunities [ 132,489]. Twitter customizes its content recommendations,
like tweets and follower suggestions, based on user behavior and preferences [490].
ByteDance has introduced innovative models for TikTok to quickly adapt recommendations to user interactions,
employing unique retrieval models and scalable systems like Monolith, which uses collisionless embedding tables for
efficient memory usage [ 491,492]. Apple has developed the Sliced Anti-symmetric Decomposition (SAD) model to
enhance collaborative filtering, allowing more nuanced user-item interactions, and explores controlled music production
using diffusion models [ 493,494]. DeepMind’s generative models improve RS by decoding Semantic IDs from user
interactions, enhancing item retrieval and system performance [495].
Table 17: Publications by Industry in Recommendation Systems
Industry Publications
E-commerce/E-Business [105, 435, 227, 303, 207, 211, 103, 181, 127, 43, 436, 437]
E-Entertainment (Music, Movies) [438, 439, 73, 440, 209, 441, 442, 432, 20, 41]
E-Health [443, 24, 81, 22, 444, 7, 445, 446, 447, 26, 448, 449, 450, 110, 451,
452, 453, 454, 455, 456, 457, 458, 459, 460, 461]
E-Government RS [115, 462, 25, 128, 463, 464, 465, 466, 467, 107, 468, 185]
E-Library and E-Learning [116, 469, 470, 177, 471, 64, 472, 473, 82]
E-Tourism/Travel [80, 21, 474, 475, 476, 62, 477, 210, 478]
E-Finance [19, 479, 480, 481, 479, 482, 123, 483]
E-News [59, 60, 152, 484, 485, 220, 192, 346, 486, 408, 487]
Miscellaneous [488, 97, 132, 489, 490, 491, 492, 493, 494, 495]
10 Discussion
10.1 Impact of this Research
This literature review have profound impacts on future research, industry practices, and collaborative endeavors. This
review article can serve many purposes within academic and professional realms. The goal of this research is beyond
merely summarizing existing knowledge but also to illuminate areas needing further investigation within RS. The
detailed summaries and tables presented in this paper can serve as educational tools that help newcomers quickly
grasp complex subjects and can be used by industry practioners to use it as a guide. Additionally, this review tracks
the development of the field, providing insights into trends and telling future directions. For example, how can the
knowledge gained through theory can be applied to address real world problems in industry.
We covered a comprehensive guide on many areas of RS, despite this, some areas and fields need more coverage, which
are briefly discussed below:
37
Page 38:
arXiv Template A P REPRINT
10.2 Limitations
Despite the rapid evolution and implementation of RS in theory across diverse sectors, current methods show several
critical limitations. Each application domain, from e-commerce to e-learning, faces unique challenges that intensify
the limitations. For instance, e-commerce RS must adapt to rapidly changing inventories and consumer trends, while
e-learning systems need to account for diverse learning styles and educational goals [ 105,236,437,28,496,473,497].
These domain-specific challenges highlight the need for RS that are not only technically robust and ethically sound but
also flexible and scalable enough to be effectively deployed by organizations of all sizes, including those with limited
resources. These limitations span various aspects of RS, including technical constraints, adaptability issues, and ethical
concerns [7, 61, 55].
Matrix factorization-based models that are considered as standard in RS theory struggle with capturing complex
user-item interactions due to inadequate latent feature representations and the inherent linearity of their interaction
models [ 4]. Neural extensions of these methods brought improvements by incorporating non-linear relationships
and capturing high-dimensional latent features [ 86,24]. However, as the volume of data grows, these deep learning-
based RS encounter their own set of challenges, particularly in maintaining computational efficiency and scalability
[498,499,500]. The substantial computational resources required for training and inference of these models pose
hurdles, especially in scenarios demanding real-time recommendations. In addition to that, many systems depend
heavily on explicit user feedback (e.g., ratings, likes), which is often sparse and not always available, neglecting implicit
feedback signals that could enhance recommendation accuracy [ 141]. Furthermore, data scarcity severely affects the
quality of recommendation systems [ 501] . Knowledge transfer from external, data-rich domains can be a solution to
enhance the modeling capabilities and performance of RS [ 501]. Additionally, approaches such as data augmentation,
self-supervised learning, and knowledge graphs can enrich data environments and sustainably address data shortages in
RS development [501].
Despite advancements, many systems still fall short in effectively integrating contextual information (e.g., time,
location) and multimodal data (e.g., text, images), limiting the depth of personalization [ 302,296,299,298]. These RS
can incorporate biases present in their training data, leading to unfair recommendations that favor certain groups or
items over others, thus raising ethical concerns [ 174,407,409,171,400,401]. Many advanced RS, especially those
based on deep learning, operate as black boxes, offering little to no insight into how recommendations are generated
[251, 502, 129]. This lack of transparency can degrade user trust and satisfaction.
The deployment of RS in real-life settings, particularly within mid to small range companies, presents additional
challenges. Limited resources and technical expertise can make the deployment of sophisticated RS challenging,
intensifying issues of scalability and adaptability to rapidly changing market conditions [ 503,336,437]. Issues with
review data, including its quality, authenticity, and the potential for manipulation, further complicate the effective
use of RS [ 45]. The extensive data collection necessary for personalized recommendations raises significant privacy
issues, particularly concerning user consent and data security. Moreover, handling user review data poses some privacy
challenges, as companies must navigate the balance between personalizing recommendations and protecting user
privacy.
10.3 Future Perspectives
Responsible AI practises RS shape user decisions, perspectives, and actions, underscoring the need for their design
to prioritize responsibility. Recent studies [ 504,505] have raised concerns about RS potential negative impacts, such
as biasing product promotions for increased profits or facilitating the spread of misinformation. Although there is a
growing interest in adopting responsible AI practices within the RS community, some challenges remain. Most existing
datasets lack comprehensive data on sensitive user attributes, complicating efforts to produce fair recommendations
[506]. Furthermore, the influence of specific model architectures on the fairness of recommendations is still not properly
understood and sparsely researched, which indicates a critical area for further investigation.
Evaluating Recommender Systems Beyond Accuracy RS are traditionally assessed using singular metrics such as
accuracy. However, this approach does not fully encapsulate the complexity of real-world user interactions. Users
demand not only precision or recall in recommendations but also need versatility and diversity in recommendations for
better user experience [ 408]. Future research should consider broadening the evaluative frameworks of RS to include
metrics that capture this diversity and serendipity.
The growing need for transparency and explainability [ 397] in RS suggests a shift towards more interpretable models
[403]. The integration of multimodal data and the application of advanced learning techniques offer promising directions
to enrich user experiences, making RS not only more effective but also more equitable and engaging. This holistic
approach will ensure that RS meet the evolving expectations of users in practical scenarios.
38
Page 39:
arXiv Template A P REPRINT
Beyond Statistical Correlations In this review, we explore the predominant focus of current RS that which involves
leveraging statistical correlations from historical user data to predict preferences and make recommendations. This
method does not explicitly determine whether one factor causes another. An active area of research is causality in RS
involves identifying how specific factors, like user behavior or item features, directly cause changes in recommendations
[507]. For instance, researchers might investigate whether increasing the exposure of action movies leads to a higher
viewership of this genre [507].
Computation and Storage Resources Despite the benefits of aggregating user-item interaction data from various users
in a central database to leverage collaborative information for recommendations, this approach has some drawbacks. It
is time-consuming, demands substantial communication and storage resources, and raises serious privacy and security
concerns. A new line of research is to perform on-device recommendation (DeviceRS) which is a small minimal
model that can be trained with lower computation and storage resources [ 508,509]. This line of research is still in its
early stages of development and deals with several open questions and challenges. Finding an efficient way to use
collaborative information from other users while keeping computation, storage, and data exposure low, and considering
the differences in data on each user’s device, remains an open challenge.
Generative AI With the rise of generative AI models like ChatGPT, researchers are exploring their potential to enhance
various fields, including RS. Conversational RS, which provide suggestions through dialogue, are gaining popularity,
and we have dedicated an entire sub-section to this exciting development. However, it is essential to emphasize that
while leveraging generative AI, we must ensure the outputs are safe and adhere to AI safety and responsible practices
[510]. This not only maximizes the benefits but also mitigates potential risks associated with these methods.
11 Conclusion
In this survey, we have reviewed the notable methodologies, applications, and challenges of RS in both academic
and industrial contexts. We proposed a framework to categorize RS publications based on modeling techniques and
their applications. The integration of RS with state-of-the-art methods such as deep learning, graph neural networks,
and LLMs demonstrates the evolution in this field and highlights its impact on improving user experiences in diverse
domains, including e-commerce, finance, media streaming, and personalized education. Despite notable advancements,
we still face challenges, including data sparsity, privacy issues, and the need for systems that are both adaptable and
explainable. This survey aims to bridge theoretical advances and algorithmic developments with practical applications,
helping the industry achieve scalability and immediate business impact. Our goal is to strengthen collaborations between
academia and industry, which is essential to translate theoretical progress into practical applications.
Acknowledgements
The authors wish to extend their profound gratitude to Scott Sanner, whose expert guidance and invaluable input were
instrumental throughout the research and writing of this survey. His deep insights into the nuances of recommender
systems significantly enhanced the quality of our work. We also appreciate the advice provided by Andres Rojas,
Ali Taiyeb, and Deval Pandya during critical stages of this project. Additionally, we thank Veronica Chatrath for her
contributions in the ideation phase. This survey could not have been accomplished without their support.
References
[1] Charu C. Aggarwal. Recommender Systems The Textbook . Number 4. Springer Cham, 2016.
[2]Shlomo Berkovsky, Tsvi Kuflik, and Francesco Ricci. Mediation of user models for enhanced personalization in
recommender systems. User Modeling and User-Adapted Interaction , 18(3), 2008.
[3]Gediminas Adomavicius and Alexander Tuzhilin. Toward the next generation of recommender systems: A
survey of the state-of-the-art and possible extensions, 2005.
[4]Francesco Ricci, Lior Rokach, and Bracha Shapira. Recommender Systems: Techniques, Applications, and
Challenges. In Recommender Systems Handbook: Third Edition , chapter Chapter 1, pages 1–35. Springer, New
York, NY , 2022.
[5]Xavier Amatriain and Justin Basilico. Past, present, and future of recommender systems: An industry perspective.
InRecSys 2016 - Proceedings of the 10th ACM Conference on Recommender Systems , 2016.
[6]Shristi Shakya Khanal, P. W.C. Prasad, Abeer Alsadoon, and Angelika Maag. A systematic review: machine
learning based recommendation systems for e-learning. Education and Information Technologies , 25(4), 2020.
39
Page 40:
arXiv Template A P REPRINT
[7]Maryam Etemadi, Sepideh Bazzaz Abkenar, Ahmad Ahmadzadeh, Mostafa Haghi Kashani, Parvaneh Asghari,
Mohammad Akbari, and Ebrahim Mahdipour. A systematic review of healthcare recommender systems: Open
issues, challenges, and techniques, 2023.
[8]Wenqi Fan, Zihuai Zhao, Jiatong Li, Yunqing Liu, Xiaowei Mei, Yiqi Wang, Zhen Wen, Fei Wang, Xiangyu
Zhao, Jiliang Tang, and Qing Li. Recommender Systems in the Era of Large Language Models (LLMs). IEEE
TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING , page 1, 2023.
[9] ACM Recommender Systems Conference. ACM RecSys Conferences, 2024.
[10] Ian MacKenzie, Chris Meyer, and Steve Noble. How Retailers can keep up with Consumers. McKinsey \&
Company - Retail Insights , (October), 2013.
[11] Mansoor Iqbal. Netflix Revenue and Usage Statistics (2024) - Business of Apps, 2024.
[12] Precision Reports. Worldwide Market Research Report, and Industry Analysis, 2024.
[13] Robin Burke, Alexander Felfernig, and Mehmet H. Göker. Recommender systems: An overview. AI Magazine ,
32(3), 2011.
[14] Xavier Amatriain and Justin Basilico. Recommender Systems in Industry: A Netflix Case Study. In Recommender
Systems Handbook , chapter Chapter 11, pages 385–419. Springer, Boston, MA, 2016.
[15] Prem Melville and Vikas Sindhwani. Recommender Systems. In Encyclopedia of Machine Learning , pages
829–838. Springer, Boston, MA, Boston, MA, 2010.
[16] Alvise De Biasio, Nicolò Navarin, and Dietmar Jannach. Economic recommender systems – a systematic review.
Electronic Commerce Research and Applications , 63:101352, 1 2024.
[17] Ankit Thakkar and Kinjal Chaudhari. A comprehensive survey on deep neural networks for stock market: The
need, challenges, and future directions, 2021.
[18] Rajat Kumar Behera, Angappa Gunasekaran, Shivam Gupta, Shampy Kamboj, and Pradip Kumar Bala. Person-
alized digital marketing recommender engine. Journal of Retailing and Consumer Services , 53, 2020.
[19] Marwa Sharaf, Ezz El Din Hemdan, Ayman El-Sayed, and Nirmeen A. El-Bahnasawy. A survey on recommen-
dation systems for financial services. Multimedia Tools and Applications , 81(12), 2022.
[20] Yashar Deldjoo, Markus Schedl, Paolo Cremonesi, and Gabriella Pasi. Recommender Systems Leveraging
Multimedia Content. ACM Computing Surveys , 53(5), 2020.
[21] Kinjal Chaudhari and Ankit Thakkar. A Comprehensive Survey on Travel Recommender Systems. Archives of
Computational Methods in Engineering , 27(5), 2020.
[22] Robin De Croon, Leen Van Houdt, Nyi Nyi Htun, Gregor Štiglic, Vero Vanden Abeele, and Katrien Verbert.
Health recommender systems: Systematic review, 2021.
[23] Zafar Ali, Yi Huang, Irfan Ullah, Junlan Feng, Chao Deng, Nimbeshaho Thierry, Asad Khan, Asim Ullah Jan,
Xiaoli Shen, Wu Rui, and Guilin Qi. Deep Learning for Medication Recommendation: A Systematic Survey,
2023.
[24] Jayita Saha, Chandreyee Chowdhury, and Suparna Biswas. Review of Machine Learning and Deep Learning
Based Recommender Systems for Health Informatics. In Deep Learning Techniques for Biomedical and Health
Informatics , chapter Chapter 6, pages 101–126. Springer, Cham, 2020.
[25] Kei Long Cheung, Dilara Durusu, Xincheng Sui, and Hein de Vries. How recommender systems could support
and enhance computer-tailored digital health programs: A scoping review. Digital health , 5, 1 2019.
[26] Yue Sun, Jia Zhou, Mengmeng Ji, Lusi Pei, and Zhiwen Wang. Development and Evaluation of Health
Recommender Systems: Systematic Scoping Review and Evidence Mapping, 2023.
[27] John K. Tarus, Zhendong Niu, and Ghulam Mustafa. Knowledge-based recommendation: a review of ontology-
based recommender systems for e-learning. Artificial Intelligence Review , 50(1), 2018.
[28] Aleksandra Klašnja-Mili ´cevi´c, Mirjana Ivanovi ´c, and Alexandros Nanopoulos. Recommender systems in e-
learning environments: a survey of the state-of-the-art and possible extensions. Artificial Intelligence Review ,
44(4), 2015.
[29] Ivens Portugal, Paulo Alencar, and Donald Cowan. The use of machine learning algorithms in recommender
systems: A systematic review, 2018.
[30] Chuan Qin, Hengshu Zhu, Fuzhen Zhuang, Qingyu Guo, Qi Zhang, Le Zhang, Chao Wang, Enhong Chen, and
Hui Xiong. A survey on knowledge graph-based recommender systems, 2020.
[31] Yongfeng Zhang and Xu Chen. Explainable recommendation: A survey and new perspectives, 2020.
40
Page 41:
arXiv Template A P REPRINT
[32] Saurabh Kulkarni and Sunil F. Rodd. Context Aware Recommendation Systems: A review of the state of the art
techniques, 2020.
[33] Shaina Raza and Chen Ding. Progress in context-aware recommender systems - An overview, 2019.
[34] Jian Wei, Jianhua He, Kai Chen, Yi Zhou, and Zuoyin Tang. Collaborative filtering and deep learning based
recommendation system for cold start items. Expert Systems with Applications , 69, 2017.
[35] Rui Chen, Qingyi Hua, Yan Shuo Chang, Bo Wang, Lei Zhang, and Xiangjie Kong. A survey of collaborative
filtering-based recommender systems: from traditional methods to hybrid methods based on social networks.
IEEE Access , 6, 2018.
[36] Erion Çano and Maurizio Morisio. Hybrid Recommender Systems: A Systematic Literature Review. Intelligent
Data Analysis , 21(6):1487–1524, 1 2019.
[37] Massimo Quadrana, Paolo Cremonesi, and Dietmar Jannach. Sequence-aware recommender systems, 2018.
[38] Shoujin Wang, Longbing Cao, Yan Wang, Quan Z. Sheng, Mehmet A. Orgun, and Defu Lian. A Survey on
Session-based Recommender Systems. ACM Computing Surveys , 54(7), 2022.
[39] Malte Ludewig, Noemi Mauro, Sara Latifi, and Dietmar Jannach. Empirical analysis of session-based recom-
mendation algorithms: A comparison of neural and non-neural approaches. User Modeling and User-Adapted
Interaction , 31(1), 2021.
[40] Dietmar Jannach, Ahtsham Manzoor, Wanling Cai, and Li Chen. A Survey on Conversational Recommender
Systems, 2021.
[41] Darius Afchar, Alessandro B. Melchiorre, Markus Schedl, Romain Hennequin, Elena V . Epure, and Manuel
Moussallam. Explainability in music recommender systems. AI Magazine , 43(2), 2022.
[42] Markus Schedl, Hamed Zamani, Ching-Wei Chen, Yashar Deldjoo, and Mehdi Elahi. Current challenges and
visions in music recommender systems research. International Journal of Multimedia Information Retrieval ,
7:95–116, 2018.
[43] M. Mehdi Afsar, Trafford Crump, and Behrouz Far. Reinforcement Learning based Recommender Systems: A
Survey. ACM Computing Surveys , 55(7), 2022.
[44] Yashar Deldjoo, Tommaso Di Noia, and Felice Antonio Merra. A Survey on Adversarial Recommender Systems:
From Attack/Defense Strategies to Generative Adversarial Networks, 2021.
[45] Mehdi Srifi, Ahmed Oussous, Ayoub Ait Lahcen, and Salma Mouline. Recommender systems based on
collaborative filtering using review texts-A survey, 2020.
[46] Shiwen Wu, Fei Sun, Wentao Zhang, Xu Xie, and Bin Cui. Graph Neural Networks in Recommender Systems:
A Survey. ACM Computing Surveys , 55(5), 2022.
[47] Chen Gao, Yu Zheng, Nian Li, Yinfeng Li, Yingrong Qin, Jinghua Piao, Yuhan Quan, Jianxin Chang, Depeng
Jin, Xiangnan He, and Yong Li. A Survey of Graph Neural Networks for Recommender Systems: Challenges,
Methods, and Directions. ACM Transactions on Recommender Systems , 1(1), 2023.
[48] Zonghan Wu, Shirui Pan, Fengwen Chen, Guodong Long, Chengqi Zhang, and Philip S. Yu. A Comprehensive
Survey on Graph Neural Networks. IEEE Transactions on Neural Networks and Learning Systems , 32(1), 2021.
[49] Zeynep Batmaz, Ali Yurekli, Alper Bilge, and Cihan Kaleli. A review on deep learning for recommender systems:
challenges and remedies. Artificial Intelligence Review , 52(1), 2019.
[50] Lei Li, Yongfeng Zhang, Dugang Liu, and Li Chen. Large Language Models for Generative Recommendation:
A Survey and Visionary Discussions. 9 2023.
[51] Likang Wu, Zhi Zheng, Zhaopeng Qiu, Hao Wang, Hongchao Gu, Tingjia Shen, Chuan Qin, Chen Zhu, Hengshu
Zhu, Qi Liu, Hui Xiong, and Enhong Chen. A Survey on Large Language Models for Recommendation. 5 2023.
[52] Benyou Wang, Qianqian Xie, Jiahuan Pei, Zhihong Chen, Prayag Tiwari, Zhao Li, and Jie Fu. Pre-trained
Language Models in Biomedical Domain: A Systematic Survey. ACM Computing Surveys , 56(3):57, 10 2021.
[53] Qingxiu Dong, Lei Li, Damai Dai, Ce Zheng, Zhiyong Wu, Baobao Chang, Xu Sun, Jingjing Xu, Lei Li, and
Zhifang Sui. A survey for in-context learning. Computer Science , 2022.
[54] Lingfei Wu, Yu Chen, Kai Shen, Xiaojie Guo, Hanning Gao, Shucheng Li, Jian Pei, and Bo Long. Graph Neural
Networks for Natural Language Processing: A Survey. Foundations and Trends in Machine Learning , 16(2),
2023.
[55] Peng Liu, Lemei Zhang, and Jon Atle Gulla. Pre-train, Prompt, and Recommendation: A Comprehensive Survey
of Language Modeling Paradigm Adaptations in Recommender Systems. Transactions of the Association for
Computational Linguistics , 11:1553–1571, 12 2023.
41
Page 42:
arXiv Template A P REPRINT
[56] Hai Ha Do, P. W.C. Prasad, Angelika Maag, and Abeer Alsadoon. Deep Learning for Aspect-Based Sentiment
Analysis: A Comparative Review, 2019.
[57] Deepjyoti Roy and Mala Dutta. A systematic review and research perspective on recommender systems. Journal
of Big Data , 9(1), 2022.
[58] Mozhgan Karimi, Dietmar Jannach, and Michael Jugovac. News recommender systems – Survey and roads
ahead. Information Processing and Management , 54(6), 2018.
[59] Shaina Raza and Chen Ding. News recommender system: a review of recent progress, challenges, and
opportunities. Artificial Intelligence Review , 55(1), 2022.
[60] Chuhan Wu, Fangzhao Wu, Yongfeng Huang, and Xing Xie. Personalized News Recommendation: Methods and
Challenges. ACM Transactions on Information Systems , 41(1), 2023.
[61] Yassine Himeur, Shahab Saquib Sohail, Faycal Bensaali, Abbes Amira, and Mamoun Alazab. Latest trends of
security and privacy in recommender systems: a comprehensive review and future perspectives. Computers &
Security , 118:102746, 2022.
[62] Joy Lal Sarkar, Abhishek Majumder, Chhabi Rani Panigrahi, Sudipta Roy, and Bibudhendu Pati. Tourism
recommendation system: a survey and future research directions. Multimedia Tools and Applications , 82(6),
2023.
[63] Eva Zangerle and Christine Bauer. Evaluating Recommender Systems: Survey and Framework. 2022.
[64] Zafar Ali, Guilin Qi, Pavlos Kefalas, Waheed Ahmad Abro, and Bahadar Ali. A graph-based taxonomy of
citation recommendation models. Artificial Intelligence Review , 53(7), 2020.
[65] Govind Kumar Jha, Manish Gaur, Preetish Ranjan, and Hardeo Kumar Thakur. A survey on trustworthy model
of recommender system. International Journal of System Assurance Engineering and Management , 14, 2023.
[66] Mario Casillo, Francesco Colace, Dajana Conte, ·Marco Lombardi, Domenico Santaniello, and Carmine
Valentino. Context-aware recommender systems and cultural heritage: a survey. Journal of Ambient Intelligence
and Humanized Computing , 14:3109–3127, 2023.
[67] Elaine Rich. User modeling via stereotypes. Cognitive Science , 3(4):329–354, 10 1979.
[68] Jussi Karlgren. An algebra for recommendations : Using reader data as a basis for measuring document proximity.
1990.
[69] Paul Resnick, Neophytos Iacovou, Mitesh Suchak, Peter Bergstrom, and John Riedl. GroupLens: An open
architecture for collaborative filtering of netnews. Proceedings of the 1994 ACM Conference on Computer
Supported Cooperative Work, CSCW 1994 , pages 175–186, 10 1994.
[70] Jonathan L. Herlocker, Joseph A. Konstan, Al Borchers, and John Riedl. An algorithmic framework for
performing collaborative filtering. In Proceedings of the 22nd Annual International ACM SIGIR Conference on
Research and Development in Information Retrieval, SIGIR 1999 , 1999.
[71] Joeran Beel Docear, Stefan Langer, Marcel Genzmehr, Bela Gipp, Corinna Breitinger, and Andreas Nürnberger.
Research Paper Recommender System Evaluation: A Quantitative Literature Survey.
[72] Yandi Xia, Giuseppe Di Fabbrizio, Shikhar Vaibhav, Ankur Datta, and Ankur 2017 Datta. A Content-based
Recommender System for E-commerce OOers and Coupons. SIGIR , 2017.
[73] How Netflix’s Recommendations System Works | Netflix Help Center.
[74] Angela Carrera-Rivera, William Ochoa, Felix Larrinaga, and Ganix Lasa. How-to conduct a systematic literature
review: A quick guide for computer science research. MethodsX , 9, 2022.
[75] ACM RecSys. Workshop on Deep Learning for Recommender Systems. In Proceedings of the 11th ACM
Conference on Recommender Systems , Como, Italy, 2017.
[76] Farah Tawfiq Abdul Hussien, Abdul Monem S. Rahma, and Hala Bahjat Abdul Wahab. Recommendation
Systems for E-commerce Systems An Overview. In Journal of Physics: Conference Series , volume 1897, 2021.
[77] Ke Chen, Beici Liang, Xiaoshuan Ma, and Minwei Gu. Learning audio embeddings with user listening data for
content-based music recommendation. In ICASSP , IEEE International Conference on Acoustics, Speech and
Signal Processing - Proceedings , volume 2021-June, 2021.
[78] Gustavo Penha and Claudia Hauff. What does BERT know about books, movies and music? Probing BERT for
Conversational Recommendation. RecSys 2020 - 14th ACM Conference on Recommender Systems , 20:388–397,
9 2020.
42
Page 43:
arXiv Template A P REPRINT
[79] Shaina Raza. Automatic fake news detection in political platforms-a transformer-based approach. In Proceedings
of the 4th Workshop on Challenges and Applications of Automated Extraction of Socio-political Events from Text
(CASE 2021) , pages 68–78, 2021.
[80] Rula A. Hamid, A. S. Albahri, Jwan K. Alwan, Z. T. Al-Qaysi, O. S. Albahri, A. A. Zaidan, Alhamzah
Alnoor, A. H. Alamoodi, and B. B. Zaidan. How smart is e-tourism? A systematic review of smart tourism
recommendation system applying data management, 2021.
[81] Thi Ngoc Trang Tran, Alexander Felfernig, Christoph Trattner, and Andreas Holzinger. Recommender systems
in the healthcare domain: state-of-the-art and research issues. Journal of Intelligent Information Systems , 57(1),
2021.
[82] Nur W. Rahayu, Ridi Ferdiana, and Sri S. Kusumawardani. A systematic review of ontology use in E-Learning
recommender system, 2022.
[83] Michael J. Pazzani and Daniel Billsus. Content-based recommendation systems. Lecture Notes in Computer
Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) , 4321
LNCS:325–341, 2007.
[84] Peter D. Turney and Patrick Pantel. From frequency to meaning: Vector space models of semantics. Journal of
Artificial Intelligence Research , 37, 2010.
[85] Alexandrin Popescul, David M Pennock, and Steve Lawrence. Probabilistic Models for Unified Collaborative
and Content-Based Recommendation in Sparse-Data Environments. Artificial Intelligence , 2001(Uai):437–444,
2001.
[86] Shuai Zhang, Lina Yao, Aixin Sun, and Yi Tay. Deep learning based recommender system: A survey and new
perspectives, 2019.
[87] Lei Zheng, Vahid Noroozi, and Philip S. Yu. Joint deep modeling of users and items using reviews for
recommendation. In WSDM 2017 - Proceedings of the 10th ACM International Conference on Web Search and
Data Mining , 2017.
[88] Chandra Bhagavatula, Sergey Feldman, Russell Power, and Waleed Ammar. Content-based citation recom-
mendation. In NAACL HLT 2018 - 2018 Conference of the North American Chapter of the Association for
Computational Linguistics: Human Language Technologies - Proceedings of the Conference , volume 1, 2018.
[89] Kevin Joseph and Hui Jiang. Content based news recommendation via shortest entity distance over knowledge
graphs. In The Web Conference 2019 - Companion of the World Wide Web Conference, WWW 2019 , 2019.
[90] Fuhu Deng, Panlong Ren, Zhen Qin, Gu Huang, and Zhiguang Qin. Leveraging Image Visual Features in
Content-Based Recommender System. Scientific Programming , 2018, 2018.
[91] Badrul Sarwar, George Karypis, Joseph Konstan, and John Riedl. Item-based collaborative filtering recom-
mendation algorithms. In Proceedings of the 10th International Conference on World Wide Web, WWW 2001 ,
2001.
[92] Yehuda Koren, Robert Bell, and Chris V olinsky. Matrix factorization techniques for recommender systems.
Computer , 42(8), 2009.
[93] Steffen Rendle. Factorization machines with libFM. ACM Transactions on Intelligent Systems and Technology ,
3(3), 2012.
[94] Xiangnan He, Lizi Liao, Hanwang Zhang, Liqiang Nie, Xia Hu, and Tat-Seng Chua. Neural collaborative
filtering. In Proceedings of the 26th international conference on world wide web , pages 173–182, 2017.
[95] Shoujin Wang, Qi Zhang, Liang Hu, Xiuzhen Zhang, Yan Wang, and Charu Aggarwal. Sequential/Session-based
Recommendations: Challenges, Approaches, Applications and Opportunities. In SIGIR 2022 - Proceedings of
the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval , 2022.
[96] Poonam B.Thorat, R. M. Goudar, and Sunita Barve. Survey on Collaborative Filtering, Content-based Filtering
and Hybrid Recommendation System. International Journal of Computer Applications , 110(4), 2015.
[97] Heng Tze Cheng, Levent Koc, Jeremiah Harmsen, Tal Shaked, Tushar Chandra, Hrishi Aradhye, Glen Anderson,
Greg Corrado, Wei Chai, Mustafa Ispir, Rohan Anil, Zakaria Haque, Lichan Hong, Vihan Jain, Xiaobing Liu, and
Hemal Shah. Wide \& deep learning for recommender systems. In ACM International Conference Proceeding
Series , volume 15-September-2016, 2016.
[98] Xiangnan He and Tat Seng Chua. Neural factorization machines for sparse predictive analytics. In SIGIR 2017 -
Proceedings of the 40th International ACM SIGIR Conference on Research and Development in Information
Retrieval , 2017.
43
Page 44:
arXiv Template A P REPRINT
[99] Huifeng Guo, Ruiming Tang, Yunming Ye, Zhenguo Li, and Xiuqiang He. DeepFM: A factorization-machine
based neural network for CTR prediction. In IJCAI International Joint Conference on Artificial Intelligence ,
volume 0, 2017.
[100] Ruoxi Wang, Gang Fu, Bin Fu, and Mingliang Wang. Deep and cross network for ad click predictions. In 2017
AdKDD and TargetAd - In conjunction with the 23rd ACM SIGKDD Conference on Knowledge Discovery and
Data Mining, KDD 2017 , 2017.
[101] Khalid Haruna, Maizatul Akmar Ismail, Abdullahi Baffa Bichi, Victor Chang, Sutrisna Wibawa, and Tutut
Herawan. A citation-based recommender system for scholarly paper recommendation. In Lecture Notes in Com-
puter Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) ,
volume 10960 LNCS, 2018.
[102] Tessy Badriyah, Erry Tri Wijayanto, Iwan Syarif, and Prima Kristalina. A hybrid recommendation system
for E-commerce based on product description and user profile. In 7th International Conference on Innovative
Computing Technology, INTECH 2017 , 2017.
[103] Mingsheng Fu, Hong Qu, Zhang Yi, Li Lu, and Yongsheng Liu. A Novel Deep Learning-Based Collaborative
Filtering Model for Recommendation System. IEEE Transactions on Cybernetics , 49(3), 2019.
[104] Sabrine Ben Abdrbbah. A Novel Recommendation Approach For Groups Based On Aggregating Top-k Lists.
Procedia Computer Science , 225:3067–3076, 1 2023.
[105] Kangning Wei, Jinghua Huang, and Shaohong Fu. A survey of E-commerce recommender systems. In
Proceedings - ICSSSM’07: 2007 International Conference on Service Systems and Service Management , 2007.
[106] Rajesh K. Jha, Sujoy Bag, Debbani Koley, Giridhar Reddy Bojja, and Subhas Barman. An appropriate and cost-
effective hospital recommender system for a patient of rural area using deep reinforcement learning. Intelligent
Systems with Applications , 18:200218, 5 2023.
[107] Omar Saeed Al-Mushayt. Automating E-Government Services with Artificial Intelligence. IEEE Access , 7,
2019.
[108] Robin Burke, Nasim Sonboli, and Aldo Ordoñez-Gauger. Balanced Neighborhoods for Multi-sided Fairness in
Recommendation. In Proceedings of Machine Learning Research , volume 81, 2018.
[109] Jiawei Chen, Hande Dong, Xiang Wang, Fuli Feng, Meng Wang, and Xiangnan He. Bias and Debias in
Recommender System: A Survey and Future Directions. ACM Transactions on Information Systems , 41(3),
2023.
[110] Sandeep K. Raghuwanshi and R. K. Pateriya. Collaborative Filtering Techniques in Recommendation Systems.
InData, Engineering and Applications , volume 1, chapter Chapter 2, pages 11–21. Springer, Singapore, 2019.
[111] Xiao Yu, Xiang Ren, Quanquan Gu, Yizhou Sun, and Jiawei Han. Collaborative Filtering with Entity Similarity
Regularization in Heterogeneous Information Networks. In Proc. of IJCAI-13 HINA workshop (IJCAI-HINA’13) ,
2013.
[112] Fuzheng Zhang, Nicholas Jing Yuan, Defu Lian, Xing Xie, and Wei Ying Ma. Collaborative knowledge base
embedding for recommender systems. In Proceedings of the ACM SIGKDD International Conference on
Knowledge Discovery and Data Mining , volume 13-17-August-2016, 2016.
[113] Hong Jian Xue, Xin Yu Dai, Jianbing Zhang, Shujian Huang, and Jiajun Chen. Deep matrix factorization models
for recommender systems. In IJCAI International Joint Conference on Artificial Intelligence , volume 0, 2017.
[114] Qiang Liu, Shu Wu, and Liang Wang. Deepstyle: Learning user preferences for visual recommendation. In
SIGIR 2017 - Proceedings of the 40th International ACM SIGIR Conference on Research and Development in
Information Retrieval , 2017.
[115] Caie Xu, Lisha Xu, Yingying Lu, Huan Xu, and Zhongliang Zhu. E-government recommendation algorithm
based on probabilistic semantic cluster analysis in combination of improved collaborative filtering in big-data
environment of government affairs. Personal and Ubiquitous Computing , 23(3-4), 2019.
[116] Martin Ebner. E-learning 2.0 = e-learning 1.0 + Web 2.0? In Proceedings - Second International Conference on
Availability, Reliability and Security, ARES 2007 , 2007.
[117] Juanhui Li, Haoyu Han, Zhikai Chen, Harry Shomer, Wei Jin, Amin Javari, and Jiliang Tang. Enhancing ID and
Text Fusion via Alternative Training in Session-based Recommendation. Proceedings of Make sure to enter the
correct conference title from your rights confirmation emai (Conference acronym ’XX) , 1, 2 2024.
[118] Luyi Ma, Nimesh Sinha, Parth Vajge, Jason H.D. Cho, Sushant Kumar, and Kannan Achan. Event-based Product
Carousel Recommendation with Query-Click Graph. In Proceedings - 2021 IEEE International Conference on
Big Data, Big Data 2021 , 2021.
44
Page 45:
arXiv Template A P REPRINT
[119] Xinshi Chen, Shuang Li, Hui Li, Shaohua Jiang, Yuan Qi, and Le Song. Generative adversarial user model for
reinforcement learning based recommendation system. In 36th International Conference on Machine Learning,
ICML 2019 , volume 2019-June, 2019.
[120] Rex Ying, Ruining He, Kaifeng Chen, Pong Eksombatchai, William L. Hamilton, and Jure Leskovec. Graph
convolutional neural networks for web-scale recommender systems. In Proceedings of the ACM SIGKDD
International Conference on Knowledge Discovery and Data Mining , 2018.
[121] Yao Cai, Fei Yu, Manish Kumar, Roderick Gladney, and Javed Mostafa. Health Recommender Systems Develop-
ment, Usage, and Evaluation from 2010 to 2022: A Scoping Review. International Journal of Environmental
Research and Public Health , 19(22), 2022.
[122] Xiao Wang, Houye Ji, Peng Cui, P. Yu, Chuan Shi, Bai Wang, and Yanfang Ye. Heterogeneous graph attention
network. In The Web Conference 2019 - Proceedings of the World Wide Web Conference, WWW 2019 , 2019.
[123] Wanying Ding, Vinay K. Chaudhri, Naren Chittar, and Krishna Konakanchi. JEL: Applying End-to-End Neural
Entity Linking in JPMorgan Chase. In 35th AAAI Conference on Artificial Intelligence, AAAI 2021 , volume 17B,
2021.
[124] K. Shailaja, B. Seetharamulu, and M. A. Jabbar. Machine Learning in Healthcare: A Review. In Proceedings
of the 2nd International Conference on Electronics, Communication and Aerospace Technology, ICECA 2018 ,
2018.
[125] Mingxiao An, Fangzhao Wu, Chuhan Wu, Kun Zhang, Zheng Liu, and Xing Xie. Neural news recommendation
with long- And short-term user representations. In ACL 2019 - 57th Annual Meeting of the Association for
Computational Linguistics, Proceedings of the Conference , 2020.
[126] Mirko Perano, Gian Luca Casali, Yulin Liu, and Tindara Abbate. Professional reviews as service: A mix method
approach to assess the value of recommender systems in the entertainment industry. Technological Forecasting
and Social Change , 169, 2021.
[127] Xiang Ma, Xiaojiang Lei, Guoshuai Zhao, and Xueming Qian. Rating prediction by exploring user’s preference
and sentiment. Multimedia Tools and Applications , 77(6), 2018.
[128] María E. Cortés-Cediel, Iván Cantador, and Olga Gil. Recommender systems for e-governance in smart cities:
State of the art and research opportunities. In ACM International Conference Proceeding Series , 2017.
[129] Yanzhang Lyu, Hongzhi Yin, Jun Liu, Mengyue Liu, Huan Liu, and Shizhuo Deng. Reliable recommendation
with review-level explanations. In Proceedings - International Conference on Data Engineering , volume
2021-April, 2021.
[130] Fajie Yuan, Alexandros Karatzoglou, Ioannis Arapakis, Joemon M. Jose, and Xiangnan He. A simple con-
volutional generative network for next item recommendation. WSDM 2019 - Proceedings of the 12th ACM
International Conference on Web Search and Data Mining , pages 582–590, 1 2019.
[131] Yu Lei, Zhitao Wang, Wenjie Li, and Hongbin Pei. Social attentive deep Q-network for recommendation. In
SIGIR 2019 - Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in
Information Retrieval , 2019.
[132] Sahin Cem Geyik, Qi Guo, Bo Hu, Cagri Ozcaglar, Ketan Thakkar, Xianren Wu, and Krishnaram Kenthapadi.
Talent search and recommendation systems at Linkedin: Practical challenges and lessons learned. In 41st
International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR 2018 ,
2018.
[133] Yilena Pérez-Almaguer, Raciel Yera, Ahmad A. Alzahrani, and Luis Martínez. Content-based group recommender
systems: A general taxonomy and further improvements. Expert Systems with Applications , 184, 2021.
[134] Chanwoo Jeong, Sion Jang, Eunjeong Park, and Sungchul Choi. A context-aware citation recommendation
model with BERT and graph convolutional networks. Scientometrics , 124(3), 2020.
[135] Srs Reddy, Sravani Nalluri, Subramanyam Kunisetti, S. Ashok, and B. Venkatesh. Content-based movie
recommendation system using genre correlation. In Smart Innovation, Systems and Technologies , volume 105,
2019.
[136] Bogdan Walek and Petra Spackova. Content-Based Recommender System for Online Stores Using Expert
System. In Proceedings - 2018 1st IEEE International Conference on Artificial Intelligence and Knowledge
Engineering, AIKE 2018 , 2018.
[137] Nagagopiraju Vullam, Sai Srinivas Vellela, Reddy Venkateswara, M. Venkateswara Rao, S. K. Khader Basha,
and D. Roja. Multi-Agent Personalized Recommendation System in E-Commerce based on User. In Proceedings
of the 2nd International Conference on Applied Artificial Intelligence and Computing, ICAAIC 2023 , 2023.
45
Page 46:
arXiv Template A P REPRINT
[138] Harsh Khatter, Shifa Arif, Utsav Singh, Sarthak Mathur, and Satvik Jain. Product Recommendation System
for E-Commerce using Collaborative Filtering and Textual Clustering. In Proceedings of the 3rd International
Conference on Inventive Research in Computing Applications, ICIRCA 2021 , 2021.
[139] Xinyan Fan, Zheng Liu, Jianxun Lian, Wayne Xin Zhao, Xing Xie, and Ji Rong Wen. Lighter and Better:
Low-Rank Decomposed Self-Attention Networks for Next-Item Recommendation. In SIGIR 2021 - Proceedings
of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval , 2021.
[140] Robin Burke. Hybrid web recommender systems. The adaptive web , pages 377–408, 2007.
[141] Jianxun Lian, Zhongxia Chen, Xiaohuan Zhou, Xing Xie, Fuzheng Zhang, and Guangzhong Sun. xDeepFM:
Combining explicit and implicit feature interactions for recommender systems. In Proceedings of the ACM
SIGKDD International Conference on Knowledge Discovery and Data Mining , 2018.
[142] Kun Zhou, Hui Yu, Wayne Xin Zhao, and Ji Rong Wen. Filter-enhanced MLP is All You Need for Sequential
Recommendation. In WWW 2022 - Proceedings of the ACM Web Conference 2022 , 2022.
[143] Kelong Mao, Jieming Zhu, Liangcai Su, Guohao Cai, Yuru Li, and Zhenhua Dong. FinalMLP: An Enhanced Two-
Stream MLP Model for CTR Prediction. In Proceedings of the 37th AAAI Conference on Artificial Intelligence,
AAAI 2023 , volume 37, 2023.
[144] Suvash Sedhain, Aditya Krishna Menony, Scott Sannery, and Lexing Xie. AutoRec: Autoencoders meet
collaborative filtering. In WWW 2015 Companion - Proceedings of the 24th International Conference on World
Wide Web , 2015.
[145] Dawen Liang, Rahul G. Krishnan, Matthew D. Hoffman, and Tony Jebara. Variational autoencoders for
collaborative filtering. In The Web Conference 2018 - Proceedings of the World Wide Web Conference, WWW
2018 , 2018.
[146] Wen Zhang, Yuhang Du, Taketoshi Yoshida, and Ye Yang. DeepRec: A deep neural network approach to
recommendation with item embedding and weighted loss function. Information Sciences , 470, 2019.
[147] Ilya Shenbin, Anton Alekseev, Elena Tutubalina, Valentin Malykh, and Sergey I. Nikolenko. RecV AE: A new
variational autoencoder for top-n recommendations with implicit feedback. In WSDM 2020 - Proceedings of the
13th International Conference on Web Search and Data Mining , 2020.
[148] Jinhyeok Park, Dain Kim, and Dongwoo Kim. Item-based variational auto-encoder for fair music recommenda-
tion. In CEUR Workshop Proceedings , volume 3318, 2022.
[149] Yaochen Zhu and Zhenzhong Chen. Variational Bandwidth Auto-Encoder for Hybrid Recommender Systems.
IEEE Transactions on Knowledge and Data Engineering , 35(5), 2023.
[150] Kamal Berahmand, Fatemeh Daneshfar, ·Elaheh, Sadat Salehi, Yuefeng Li, Yue Xu, Elaheh Sadat Salehi, and
K Berahmand. Autoencoders and their applications in machine learning: a survey. Artificial Intelligence Review ,
123.
[151] A. Razia Sulthana, Maulika Gupta, Shruthi Subramanian, and Sakina Mirza. Improvising the performance of
image-based recommendation system using convolution neural networks and deep learning. Soft Computing ,
24(19):14531–14544, 10 2020.
[152] Hongwei Wang, Fuzheng Zhang, Xing Xie, and Minyi Guo. DKN: Deep knowledge-aware network for news
recommendation. In The Web Conference 2018 - Proceedings of the World Wide Web Conference, WWW 2018 ,
2018.
[153] Jordi Pons and Xavier Serra. musicnn: Pre-trained convolutional neural networks for music audio tagging.
Papers with Code , 9 2019.
[154] Ming Chen, Tianyi Ma, and Xiuze Zhou. CoCNN: Co-occurrence CNN for recommendation. Expert Systems
with Applications , 195, 2022.
[155] Yu Wang, Yuying Zhao, Yi Zhang, and Tyler Derr. Collaboration-Aware Graph Convolutional Network for
Recommender Systems. ACM Web Conference 2023 - Proceedings of the World Wide Web Conference, WWW
2023 , pages 91–101, 4 2023.
[156] Laith Alzubaidi, Jinglan Zhang, Amjad J. Humaidi, Ayad Al-Dujaili, Ye Duan, Omran Al-Shamma, J. Santamaría,
Mohammed A. Fadhel, Muthana Al-Amidie, and Laith Farhan. Review of deep learning: concepts, CNN
architectures, challenges, applications, future directions. Journal of Big Data 2021 8:1 , 8(1):1–74, 3 2021.
[157] Zhu Sun, Jie Yang, Jie Zhang, Alessandro Bozzon, Long Kai Huang, and Chi Xu. Recurrent knowledge graph
embedding for effective recommendation. In RecSys 2018 - 12th ACM Conference on Recommender Systems ,
2018.
46
Page 47:
arXiv Template A P REPRINT
[158] Balázs Hidasi, Alexandros Karatzoglou, Linas Baltrunas, and Domonkos Tikk. Session-based recommendations
with recurrent neural networks. In 4th International Conference on Learning Representations, ICLR 2016 -
Conference Track Proceedings , 2016.
[159] Balázs Hidasi and Alexandros Karatzoglou. Recurrent neural networks with top-k gains for session-based
recommendations. In Proceedings of the 27th ACM international conference on information and knowledge
management , pages 843–852, 2018.
[160] Jing Li, Pengjie Ren, Zhumin Chen, Zhaochun Ren, Tao Lian, and Jun Ma. Neural attentive session-based
recommendation. In International Conference on Information and Knowledge Management, Proceedings ,
volume Part F131841, 2017.
[161] Wang Cheng Kang and Julian McAuley. Self-Attentive Sequential Recommendation. In Proceedings - IEEE
International Conference on Data Mining, ICDM , volume 2018-November, 2018.
[162] Qiannan Zhu, Xiaofei Zhou, Zeliang Song, Jianlong Tan, and Li Guo. DAN: Deep attention neural network
for news recommendation. In 33rd AAAI Conference on Artificial Intelligence, AAAI 2019, 31st Innovative
Applications of Artificial Intelligence Conference, IAAI 2019 and the 9th AAAI Symposium on Educational
Advances in Artificial Intelligence, EAAI 2019 , 2019.
[163] Yufei Feng, Fuyu Lv, Weichen Shen, Menghan Wang, Fei Sun, Yu Zhu, and Keping Yang. Deep session interest
network for click-through rate prediction. In IJCAI International Joint Conference on Artificial Intelligence ,
volume 2019-August, 2019.
[164] Shaina Raza and Chen Ding. Deep Neural Network to Tradeoff between Accuracy and Diversity in a News
Recommender System. Proceedings - 2021 IEEE International Conference on Big Data, Big Data 2021 , pages
5246–5256, 2021.
[165] Ngo Xuan Bach, Dang Hoang Long, and Tu Minh Phuong. Recurrent convolutional networks for session-based
recommendations. Neurocomputing , 411, 2020.
[166] Xiaokang Zhou, Yue Li, and Wei Liang. CNN-RNN Based Intelligent Recommendation for Online Medical
Pre-Diagnosis Support. IEEE/ACM Transactions on Computational Biology and Bioinformatics , 18(3):912–921,
5 2021.
[167] Lei Guo, Jinyu Zhang, Tong Chen, Xinhua Wang, and Hongzhi Yin. Reinforcement Learning-enhanced Shared-
account Cross-domain Sequential Recommendation. IEEE TRANSACTIONS ON KNOWLEDGE AND DATA
ENGINEERING , XX:1, 2022.
[168] Na Zhao, Zhen Long, Jian Wang, and Zhi Dan Zhao. AGRE: A knowledge graph recommendation algorithm
based on multiple paths embeddings RNN encoder. Knowledge-Based Systems , 259, 2023.
[169] Zhiwei Guo and Heng Wang. A Deep Graph Neural Network-Based Mechanism for Social Recommendations.
IEEE Transactions on Industrial Informatics , 17(4), 2021.
[170] Ruiping Yin, Kan Li, Guangquan Zhang, and Jie Lu. A deeper graph neural network for recommender systems.
Knowledge-Based Systems , 185, 2019.
[171] Masoud Mansoury, Himan Abdollahpouri, Mykola Pechenizkiy, Bamshad Mobasher, and Robin Burke. A
Graph-Based Approach for Mitigating Multi-Sided Exposure Bias in Recommender Systems. ACM Transactions
on Information Systems , 40(2), 2022.
[172] Siwei Liu, Iadh Ounis, Craig MacDonald, and Zaiqiao Meng. A Heterogeneous Graph Neural Model for
Cold-start Recommendation. In SIGIR 2020 - Proceedings of the 43rd International ACM SIGIR Conference on
Research and Development in Information Retrieval , 2020.
[173] Xiaohan Zou. A Survey on Application of Knowledge Graph. In Journal of Physics: Conference Series , volume
1487, 2020.
[174] Michael D Ekstrand, Mucun Tian, Jennifer D Ekstrand, Oghenemaro Anuyah, David Mcneill, and Maria Soledad
Pera. All The Cool Kids, How Do They Fit In? Popularity and Demographic Biases in Recommender Evaluation
and Effectiveness. Proceedings of Machine Learning Research , 81, 2018.
[175] Bin Liang, Hang Su, Lin Gui, Erik Cambria, and Ruifeng Xu. Aspect-based sentiment analysis via affective
knowledge enhanced graph convolutional networks. Knowledge-Based Systems , 235, 2022.
[176] Yuting Ye, Jingren Zhou, Xuwu Wang, Yanghua Xiao, Jiangchao Yao, Kunyang Jia, and Hongxia Yang. Bayes
Embedding (BEM): Refining representation by integrating knowledge graphs and behavior-specific networks. In
International Conference on Information and Knowledge Management, Proceedings , 2019.
[177] Xiaoyan Cai, Yu Zheng, Libin Yang, Tao Dai, and Lantian Guo. Bibliographic Network Representation Based
Personalized Citation Recommendation. IEEE Access , 7, 2019.
47
Page 48:
arXiv Template A P REPRINT
[178] Avishek Joey Bose and William L. Hamilton. Compositional fairness constraints for graph embeddings. In 36th
International Conference on Machine Learning, ICML 2019 , volume 2019-June, 2019.
[179] Mengqi Zhang, Shu Wu, Xueli Yu, Qiang Liu, and Liang Wang. Dynamic Graph Neural Networks for Sequential
Recommendation. IEEE Transactions on Knowledge and Data Engineering , 35(5), 2023.
[180] Xiang Wang, Dingxian Wang, Canran Xu, Xiangnan He, Yixin Cao, and Tat Seng Chua. Explainable reasoning
over knowledge graphs for recommendation. In 33rd AAAI Conference on Artificial Intelligence, AAAI 2019,
31st Innovative Applications of Artificial Intelligence Conference, IAAI 2019 and the 9th AAAI Symposium on
Educational Advances in Artificial Intelligence, EAAI 2019 , 2019.
[181] Ruihong Qiu, Zi Huang, Jingjing Li, and Hongzhi Yin. Exploiting Cross-session Information for Session-based
Recommendation with Graph Neural Networks. ACM Transactions on Information Systems , 38(3), 2020.
[182] Tao Dai, Li Zhu, Xiaoyan Cai, Shirui Pan, and Sheng Yuan. Explore semantic topics and author communities for
citation recommendation in bipartite bibliographic network. Journal of Ambient Intelligence and Humanized
Computing , 9(4), 2018.
[183] Zuohui Fu, Yikun Xian, Ruoyuan Gao, Jieyu Zhao, Qiaoying Huang, Yingqiang Ge, Shuyuan Xu, Shijie Geng,
Chirag Shah, Yongfeng Zhang, and Gerard De Melo. Fairness-Aware Explainable Recommendation over
Knowledge Graphs. In SIGIR 2020 - Proceedings of the 43rd International ACM SIGIR Conference on Research
and Development in Information Retrieval , 2020.
[184] Zekun Li, Zeyu Cui, Shu Wu, Xiaoyu Zhang, and Liang Wang. Fi-GNN: Modeling feature interactions via graph
neural networks for CTR prediction. In International Conference on Information and Knowledge Management,
Proceedings , 2019.
[185] Lingyuan Kong, Hao Ding, and Guangwei Hu. GCNSLIM: Graph Convolutional Network with Sparse Linear
Methods for E-government Service Recommendation. Knowledge-Based Systems , V olume 292, 5 2023.
[186] Ziyang Wang, Wei Wei, Gao Cong, Xiao Li Li, Xian Ling Mao, and Minghui Qiu. Global Context Enhanced
Graph Neural Networks for Session-based Recommendation. In SIGIR 2020 - Proceedings of the 43rd Interna-
tional ACM SIGIR Conference on Research and Development in Information Retrieval , 2020.
[187] Chun Liu, Yuxiang Li, Hong Lin, and Chaojie Zhang. GNNRec: gated graph neural network for session-based
social recommendation model. Journal of Intelligent Information Systems , 60(1), 2023.
[188] Jiancan Wu, Xiangnan He, Xiang Wang, Qifan Wang, Weijian Chen, Jianxun Lian, and Xing Xie. Graph
convolution machine for context-aware recommender system. Frontiers of Computer Science , 16(6), 2022.
[189] Van Den Rianne Berg, Thomas N Kipf, and Max Welling. Graph Convolutional Matrix Completion. Papers with
Code , 2017.
[190] Qiang He, Xinkai Li, and Biao Cai. Graph neural network recommendation algorithm based on improved dual
tower model. Scientific Reports 2024 14:1 , 14(1):1–13, 2 2024.
[191] Wenqi Fan, Yao Ma, Qing Li, Yuan He, Eric Zhao, Jiliang Tang, and Dawei Yin. Graph neural networks for
social recommendation. In The Web Conference 2019 - Proceedings of the World Wide Web Conference, WWW
2019 , 2019.
[192] Zhaopeng Qiu, Yunfan Hu, and Xian Wu. Graph Neural News Recommendation with User Existing and Potential
Interest Modeling. ACM Transactions on Knowledge Discovery from Data , 16(5), 2022.
[193] Meng Liu, Jianjun Li, Ke Liu, Chaoyang Wang, Pan Peng, Guohui Li, Yongjing Cheng, Guohui Jia, and Wei Xie.
Graph-ICF: Item-based collaborative filtering based on graph neural network. Knowledge-Based Systems , 251,
2022.
[194] William L Hamilton, Rex Ying, and Jure Leskovec. Inductive Representation Learning on Large Graphs.
Advances in Neural Information Processing Systems , 2017.
[195] Jun Zhao, Zhou Zhou, Wei Ning, Ziyu Guan, Guang Qiu, Wei Zhao, and Xiaofei He. IntentGC: A scalable
graph convolution framework fusing heterogeneous information for recommendation. In Proceedings of the
ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , 2019.
[196] Weizhi Ma, Woojeong Jin, Min Zhang, Chenyang Wang, Yue Cao, Yiqun Liu, Shaoping Ma, and Xiang Ren.
Jointly learning explainable rules for recommendation with knowledge graph. In The Web Conference 2019 -
Proceedings of the World Wide Web Conference, WWW 2019 , 2019.
[197] Xiang Wang, Xiangnan He, Yixin Cao, Meng Liu, and Tat Seng Chua. KGAT: Knowledge graph attention
network for recommendation. In Proceedings of the ACM SIGKDD International Conference on Knowledge
Discovery and Data Mining , 2019.
48
Page 49:
arXiv Template A P REPRINT
[198] Guoliang Ji, Shizhu He, Liheng Xu, Kang Liu, and Jun Zhao. Knowledge graph embedding via dynamic mapping
matrix. In ACL-IJCNLP 2015 - 53rd Annual Meeting of the Association for Computational Linguistics and
the 7th International Joint Conference on Natural Language Processing of the Asian Federation of Natural
Language Processing, Proceedings of the Conference , volume 1, 2015.
[199] Heiko Paulheim. Knowledge graph refinement: A survey of approaches and evaluation methods. Semantic Web ,
8(3), 2017.
[200] Le Wu, Lei Chen, Pengyang Shao, Richang Hong, Xiting Wang, and Meng Wang. Learning fair representations
for recommendation: A graph-based perspective. In The Web Conference 2021 - Proceedings of the World Wide
Web Conference, WWW 2021 , 2021.
[201] Alberto García-Durán and Mathias Niepert. Learning graph representations with embedding propagation. In
Advances in Neural Information Processing Systems , volume 2017-December, 2017.
[202] Xiangnan He, Kuan Deng, Xiang Wang, Yan Li, Yong Dong Zhang, and Meng Wang. LightGCN: Simplifying
and Powering Graph Convolution Network for Recommendation. In SIGIR 2020 - Proceedings of the 43rd
International ACM SIGIR Conference on Research and Development in Information Retrieval , 2020.
[203] Tinglin Huang, Yuxiao Dong, Ming Ding, Zhen Yang, Wenzheng Feng, Xinyu Wang, and Jie Tang. MixGCF:
An Improved Training Method for Graph Neural Network-based Recommender Systems. In Proceedings of the
ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , 2021.
[204] Yinwei Wei, Xiangnan He, Xiang Wang, Richang Hong, Liqiang Nie, and Tat Seng Chua. MMGCN: Multi-modal
graph convolution network for personalized recommendation of micro-video. In MM 2019 - Proceedings of the
27th ACM International Conference on Multimedia , 2019.
[205] Ke Wang, Yanmin Zhu, Tianzi Zang, Chunyang Wang, Kuan Liu, and Peibo Ma. Multi-aspect Graph Contrastive
Learning for Review-enhanced Recommendation. ACM Transactions on Information Systems , 42(2), 2023.
[206] Hongwei Wang, Fuzheng Zhang, Miao Zhao, Wenjie Li, Xing Xie, and Minyi Guo. Multi-task feature learning
for knowledge graph enhanced recommendation. In The Web Conference 2019 - Proceedings of the World Wide
Web Conference, WWW 2019 , 2019.
[207] Mingsong Mao, Jie Lu, Jialin Han, and Guangquan Zhang. Multiobjective e-commerce recommendations based
on hypergraph ranking. Information Sciences , 471, 2019.
[208] Xiang Wang, Xiangnan He, Meng Wang, Fuli Feng, and Tat Seng Chua. Neural graph collaborative filtering.
SIGIR 2019 - Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in
Information Retrieval , pages 165–174, 7 2019.
[209] Marco De Nadai, Francesco Fabbri, Paul Gigioli, Alice Wang, Ang Li, Fabrizio Silvestri, Laura Kim, Shawn Lin,
Vladan Radosavljevic, Sandeep Ghael, David Nyhan, Hugues Bouchard, Mounia Lalmas-Roelleke, Andreas
Damianou, Fab-Rizio Silvestri, and Andreas Dami. Personalized Audiobook Recommendations at Spotify
Through Graph Neural Networks. WWW ’24 Companion, May 13â•fi17, 2024, Singapore , 1, 3 2024.
[210] Soheil Rezaee, Abolghasem Sadeghi-Niaraki, Maryam Shakeri, and Soo Mi Choi. Personalized augmented
reality based tourism system: Big data and user demographic contexts. Applied Sciences (Switzerland) , 11(13),
2021.
[211] Shakila Shaikh, Sheetal Rathi, and Prachi Janrao. Recommendation system in E-Commerce Websites: A graph
based approached. In Proceedings - 7th IEEE International Advanced Computing Conference, IACC 2017 , 2017.
[212] Xiang Wang, Yaokun Xu, Xiangnan He, Yixin Cao, Meng Wang, and Tat Seng Chua. Reinforced Negative
Sampling over Knowledge Graph for Recommendation. The Web Conference 2020 - Proceedings of the World
Wide Web Conference, WWW 2020 , pages 99–109, 4 2020.
[213] Yikun Xian, Zuohui Fu, S. Muthukrishnan, Gerard De Melo, and Yongfeng Zhang. Reinforcement knowledge
graph reasoning for explainable recommendation. In SIGIR 2019 - Proceedings of the 42nd International ACM
SIGIR Conference on Research and Development in Information Retrieval , 2019.
[214] Yu Lei, Hongbin Pei, Hanqi Yan, and Wenjie Li. Reinforcement Learning based Recommendation with Graph
Convolutional Q-network. In SIGIR 2020 - Proceedings of the 43rd International ACM SIGIR Conference on
Research and Development in Information Retrieval , 2020.
[215] Hongwei Wang, Fuzheng Zhang, Jialin Wang, Miao Zhao, Wenjie Li, Xing Xie, and Minyi Guo. RippleNet:
Propagating user preferences on the knowledge graph for recommender systems. In International Conference on
Information and Knowledge Management, Proceedings , 2018.
49
Page 50:
arXiv Template A P REPRINT
[216] Shu Wu, Yuyuan Tang, Yanqiao Zhu, Liang Wang, Xing Xie, and Tieniu Tan. Session-based recommendation
with graph neural networks. In 33rd AAAI Conference on Artificial Intelligence, AAAI 2019, 31st Innovative
Applications of Artificial Intelligence Conference, IAAI 2019 and the 9th AAAI Symposium on Educational
Advances in Artificial Intelligence, EAAI 2019 , 2019.
[217] Chaoyang Wang, Zhiqiang Guo, Peng Pan, Jianjun Li, Guohui Li, Z Guo, J Li, G Li, and P Pan. A Text-
based Deep Reinforcement Learning Framework Using Self-supervised Graph Representation for Interactive
Recommendation. ACM/IMS Transactions on Data Science (TDS) , 2(4):1–25, 5 2022.
[218] Xiang Chen, Ningyu Zhang, Lei Li, Shumin Deng, Chuanqi Tan, Changliang Xu, Fei Huang, Luo Si, and
Huajun Chen. Hybrid Transformer with Multi-level Fusion for Multimodal Knowledge Graph Completion. In
SIGIR 2022 - Proceedings of the 45th International ACM SIGIR Conference on Research and Development in
Information Retrieval , 2022.
[219] Tajuddeen Rabiu Gwadabe and Ying Liu. Improving graph neural network for session-based recommendation
system via non-sequential interactions. Neurocomputing , 468:111–122, 1 2022.
[220] Zhenyan Ji, Mengdan Wu, Hong Yang, and José Enrique Armendáriz Íñigo. Temporal sensitive heterogeneous
graph neural network for news recommendation. Future Generation Computer Systems , 125, 2021.
[221] Yixin Cao, Xiang Wang, Xiangnan He, Zikun Hu, and Tat Seng Chua. Unifying knowledge graph learning and
recommendation: Towards a better understanding of user preferences. In The Web Conference 2019 - Proceedings
of the World Wide Web Conference, WWW 2019 , 2019.
[222] Junliang Yu, Xin Xia, Tong Chen, Lizhen Cui, Nguyen Quoc Viet Hung, and Hongzhi Yin. XSimGCL: Towards
Extremely Simple Graph Contrastive Learning for Recommendation. IEEE Transactions on Knowledge and
Data Engineering , 36(2), 2024.
[223] Sen Zhao, Wei Wei, Yifan Liu, Ziyang Wang, Wendi Li, Xian-Ling Mao, Shuai Zhu, Minghui Yang, and
Zujie Wen. Towards Hierarchical Policy Learning for Conversational Recommendation with Hypergraph-based
Reinforcement Learning. 2023.
[224] He Zhang, Bang Wu, Xingliang Yuan, Shirui Pan, Hanghang Tong, and Jian Pei. Trustworthy Graph Neural
Networks: Aspects, Methods and Trends. 5 2022.
[225] Maral Kolahkaj, Ali Harounabadi, Alireza Nikravanshalmani, and Rahim Chinipardaz. A hybrid context-aware
approach for e-tourism package recommendation based on asymmetric similarity measurement and sequential
pattern mining. Electronic Commerce Research and Applications , 42, 2020.
[226] Weihua Yuan, Hong Wang, Xiaomei Yu, Nan Liu, and Zhenghao Li. Attention-based context-aware sequential
recommendation model. Information Sciences , 510, 2019.
[227] Qiwei Chen, Huan Zhao, Wei Li, Pipei Huang, and Wenwu Ou. Behavior sequence transformer for E-commerce
recommendation in Alibaba. In Proceedings of the ACM SIGKDD International Conference on Knowledge
Discovery and Data Mining , 2019.
[228] Fei Sun, Jun Liu, Jian Wu, Changhua Pei, Xiao Lin, Wenwu Ou, and Peng Jiang. Bert4rec: Sequential
recommendation with bidirectional encoder representations from transformer. In International Conference on
Information and Knowledge Management, Proceedings , 2019.
[229] Jin Huang, Wayne Xin Zhao, Hongjian Dou, Ji Rong Wen, and Edward Y . Chang. Improving sequential
recommendation with knowledge-enhanced memory networks. In 41st International ACM SIGIR Conference on
Research and Development in Information Retrieval, SIGIR 2018 , 2018.
[230] Pengfei Wang, Yu Fan, Long Xia, Wayne Xin Zhao, Shaozhang Niu, and Jimmy Huang. KERL: A Knowledge-
Guided Reinforcement Learning Model for Sequential Recommendation. In SIGIR 2020 - Proceedings of the
43rd International ACM SIGIR Conference on Research and Development in Information Retrieval , 2020.
[231] Balázs Hidasi and Alexandros Karatzoglou. Recurrent neural networks with top-k gains for session-based
recommendations. In Proceedings of the 27th ACM international conference on information and knowledge
management , pages 843–852, 2018.
[232] Pengjie Ren, Zhumin Chen, Jing Li, Zhaochun Ren, Jun Ma, and Maarten de Rijke. RepeatNet: A repeat aware
neural recommendation machine for session-based recommendation. In 33rd AAAI Conference on Artificial
Intelligence, AAAI 2019, 31st Innovative Applications of Artificial Intelligence Conference, IAAI 2019 and the
9th AAAI Symposium on Educational Advances in Artificial Intelligence, EAAI 2019 , 2019.
[233] Matthew E. Peters, Waleed Ammar, Chandra Bhagavatula, and Russell Power. Semi-supervised sequence tagging
with bidirectional language models. In ACL 2017 - 55th Annual Meeting of the Association for Computational
Linguistics, Proceedings of the Conference (Long Papers) , volume 1, 2017.
50
Page 51:
arXiv Template A P REPRINT
[234] Shoujin Wang, Liang Hu, Yan Wang, Longbing Cao, Quan Z. Sheng, and Mehmet Orgun. Sequential recom-
mender systems: Challenges, progress and prospects. In IJCAI International Joint Conference on Artificial
Intelligence , volume 2019-August, 2019.
[235] Chen Wu and Ming Yan. Session-aware information embedding for E-commerce product recommendation. In
International Conference on Information and Knowledge Management, Proceedings , volume Part F131841,
2017.
[236] Dietmar Jannach, Malte Ludewig, and Lukas Lerche. Session-based item recommendation in e-commerce: on
short-term intents, reminders, trends and discounts. User Modeling and User-Adapted Interaction , 27(3-5), 2017.
[237] Tipajin Thaipisutikul and Ying Nong Chen. An improved deep sequential model for context-aware POI
recommendation. Multimedia Tools and Applications , 83(1), 2024.
[238] Tipajin Thaipisutikul and Ying Nong Chen. An improved deep sequential model for context-aware POI
recommendation. Multimedia Tools and Applications , 83(1), 2024.
[239] Yupeng Hou, Binbin Hu, Zhiqiang Zhang, and Wayne Xin Zhao. CORE: Simple and Effective Session-based
Recommendation within Consistent Representation Space. In SIGIR 2022 - Proceedings of the 45th International
ACM SIGIR Conference on Research and Development in Information Retrieval , 2022.
[240] Xinyu Du, Huanhuan Yuan, Pengpeng Zhao, Jianfeng Qu, Fuzhen Zhuang, Guanfeng Liu, Yanchi Liu, and
Victor S. Sheng. Frequency Enhanced Hybrid Attention Network for Sequential Recommendation. In SIGIR 2023
- Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information
Retrieval , 2023.
[241] Jianyang Zhai, Xiawu Zheng, Chang Dong Wang, Hui Li, and Yonghong Tian. Knowledge Prompt-tuning
for Sequential Recommendation. In MM 2023 - Proceedings of the 31st ACM International Conference on
Multimedia , 2023.
[242] Jianyang Zhai, Xiawu Zheng, Chang Dong Wang, Hui Li, and Yonghong Tian. Knowledge Prompt-tuning
for Sequential Recommendation. In MM 2023 - Proceedings of the 31st ACM International Conference on
Multimedia , 2023.
[243] Jinpeng Wang, Ziyun Zeng, Yunxiao Wang, Yuting Wang, Xingyu Lu, Tianxiang Li, Jun Yuan, Rui Zhang,
Hai Tao Zheng, and Shu Tao Xia. MISSRec: Pre-training and Transferring Multi-modal Interest-aware Sequence
Representation for Recommendation. In MM 2023 - Proceedings of the 31st ACM International Conference on
Multimedia , 2023.
[244] Wei Ji, Xiangyan Liu, An Zhang, Yinwei Wei, Yongxin Ni, and Xiang Wang. Online Distillation-enhanced Multi-
modal Transformer for Sequential Recommendation. In MM 2023 - Proceedings of the 31st ACM International
Conference on Multimedia , 2023.
[245] Jiaxi Tang and Ke Wang. Personalized top-N sequential recommendation via convolutional sequence embedding.
InWSDM 2018 - Proceedings of the 11th ACM International Conference on Web Search and Data Mining ,
volume 2018-Febuary, 2018.
[246] Balázs Hidasi and Alexandros Karatzoglou. Recurrent Neural Net-works with Top-k Gains for Session-based
Recommendations. page 10, 2018.
[247] Gabriel De Souza Pereira Moreira, Sara Rabhi, Jeong Min Lee, Ronay Ak, and Even Oldridge. Transformers4Rec:
Bridging the Gap between NLP and sequential/session-based recommendation. In RecSys 2021 - 15th ACM
Conference on Recommender Systems , 2021.
[248] Liwei Wu, Shuqing Li, Cho-Jui Hsieh, and James Sharpnack. SSE-PT: Sequential recommendation via per-
sonalized transformer. In Proceedings of the 14th ACM conference on recommender systems , pages 328–337,
2020.
[249] Hengchang Hu, Wei Guo, Yong Liu, and Min Yen Kan. Adaptive Multi-Modalities Fusion in Sequential
Recommendation Systems. International Conference on Information and Knowledge Management, Proceedings ,
pages 843–853, 10 2023.
[250] Liwei Huang, Mingsheng Fu, Fan Li, Hong Qu, Yangjun Liu, and Wenyu Chen. A deep reinforcement learning
based long-term recommender system. Knowledge-Based Systems , 213, 2021.
[251] Xiting Wang, Yiru Chen, Jie Yang, Le Wu, Zhengtao Wu, and Xing Xie. A Reinforcement Learning Framework
for Explainable Recommendation. In Proceedings - IEEE International Conference on Data Mining, ICDM ,
volume 2018-November, 2018.
[252] Xiangyu Zhao, Liang Zhang, Zhuoye Ding, Dawei Yin, Yihong Zhao, and Jiliang Tang. Deep Reinforcement
Learning for List-wise Recommendations. CEUR Workshop Proceedings , 1828, 2017.
51
Page 52:
arXiv Template A P REPRINT
[253] Xiangyu Zhao, Long Xia, Liang Zhang, Zhuoye Ding, Dawei Yin, and Jiliang Tang. Deep reinforcement learning
for page-wise recommendations. In RecSys 2018 - 12th ACM Conference on Recommender Systems , 2018.
[254] Xiaocong Chen, Lina Yao, Julian McAuley, Guanglin Zhou, and Xianzhi Wang. Deep reinforcement learning in
recommender systems: A survey and new perspectives. Knowledge-Based Systems , 264, 2023.
[255] Hado Van Hasselt, Arthur Guez, and David Silver. Deep reinforcement learning with double Q-Learning. In
30th AAAI Conference on Artificial Intelligence, AAAI 2016 , 2016.
[256] Kai Arulkumaran, Marc Peter Deisenroth, Miles Brundage, and Anil Anthony Bharath. Deep reinforcement
learning: A brief survey, 2017.
[257] Marios Kokkodis and Panagiotis G. Ipeirotis. Demand-aware career path recommendations: A reinforcement
learning approach. Management Science , 67(7), 2021.
[258] Rong Gao, Haifeng Xia, Jing Li, Donghua Liu, Shuai Chen, and Gang Chun. DRCGR: Deep reinforcement
learning framework incorporating CNN and GAN-Based for interactive recommendation. Proceedings - IEEE
International Conference on Data Mining, ICDM , 2019-November:1048–1053, 11 2019.
[259] Guanjie Zheng, Fuzheng Zhang, Zihan Zheng, Yang Xiang, Nicholas Jing Yuan, Xing Xie, and Zhenhui Li.
DRN: A deep reinforcement learning framework for news recommendation. In The Web Conference 2018 -
Proceedings of the World Wide Web Conference, WWW 2018 , 2018.
[260] Yang Zhang, Chenwei Zhang, and Xiaozhong Liu. Dynamic scholarly collaborator recommendation via
competitive multi-agent reinforcement learning. In RecSys 2017 - Proceedings of the 11th ACM Conference on
Recommender Systems , 2017.
[261] Yuanqing Yu, Chongming Gao, Jiawei Chen, Heng Tang, Yuefeng Sun, Qian Chen, Weizhi Ma, and Min
Zhang. EasyRL4Rec: A User-Friendly Code Library for Reinforcement Learning Based Recommender Systems.
Proceedings of Make sure to enter the correct conference title from your rights confirmation emai (Conference
acronym ’XX) , 1, 2 2024.
[262] Xu He, Bo An, Yanghua Li, Haikai Chen, Rundong Wang, Xinrun Wang, Runsheng Yu, Xin Li, and Zhirong
Wang. Learning to Collaborate in Multi-Module Recommendation via Multi-Agent Reinforcement Learning
without Communication. In RecSys 2020 - 14th ACM Conference on Recommender Systems , 2020.
[263] Krishna Sai Gottipati, Yashaswi Pathak, Rohan Nuttall, Raviteja Chunduru, Ahmed Touati, Sriram Ganap-
athi Subramanian, Matthew E Taylor, and Sarath Chandar. Maximum Reward Formulation In Reinforcement
Learning. Papers with Code , 2023.
[264] Tao Gui, Peng Liu, Qi Zhang, Liang Zhu, Minlong Peng, Yunhua Zhou, and Xuanjing Huang. Mention
recommendation in twitter with cooperative multi-agent reinforcement learning. In SIGIR 2019 - Proceedings of
the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval , 2019.
[265] Xueying Bai, Jian Guan, and Hongning Wang. Model-based reinforcement learning with adversarial training for
online recommendation. In Advances in Neural Information Processing Systems , volume 32, 2019.
[266] Qihua Zhang, Junning Liu, Yuzhuo Dai, Yiyan Qi, Yifan Yuan, Kunlun Zheng, Fan Huang, and Xianfeng Tan.
Multi-Task Fusion via Reinforcement Learning for Long-Term User Satisfaction in Recommender Systems. In
Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , 2022.
[267] Jia Wei Chang, Ching Yi Chiou, Jia Yi Liao, Ying Kai Hung, Chien Che Huang, Kuan Cheng Lin, and Ying Hung
Pu. Music recommender using deep embedding-based features and behavior-based reinforcement learning.
Multimedia Tools and Applications , 80(26-27), 2021.
[268] Zhiyuan Wang and Wenjie Wang. Parameter design of grid-tied inverter using reinforcement learning. IET
Conference Proceedings , 2021(9):116–120, 2021.
[269] Binbin Hu, Chuan Shi, and Jian Liu. Playlist recommendation based on reinforcement learning. In IFIP Advances
in Information and Communication Technology , volume 510, 2017.
[270] David Rohde, Stephen Bonner, Travis Dunlop, Flavian Vasile, and Alexandros Karatzoglou. RecoGym: A
Reinforcement Learning Environment for the problem of Product Recommendation in Online Advertising. ACM
Reference Format: RecSys , 18, 8 2018.
[271] Sungwoon Choi, Heonseok Ha, Uiwon Hwang, Chanju Kim, Jung-Woo Ha, and Sungroh Yoon. Reinforcement
Learning based Recommender System using Biclustering Technique. Computing Research Repository (CoRR) , 1
2018.
[272] Wacharawan Intayoad, Chayapol Kamyod, and Punnarumol Temdee. Reinforcement Learning for Online
Learning Recommendation System. In 6th Global Wireless Summit, GWS 2018 , 2018.
52
Page 53:
arXiv Template A P REPRINT
[273] Eugene Ie, Vihan Jain, Jing Wang, Sanmit Narvekar, Ritesh Agarwal, Rui Wu, Heng-Tze Cheng, Morgane
Lustman, Vince Gatto, Paul Covington, Jim McFadden, Tushar Chandra, and Craig Boutilier. Reinforcement
Learning for Slate-based Recommender Systems: A Tractable Decomposition and Practical Methodology. arXiv ,
5 2019.
[274] Yujing Hu, Qing Da, Anxiang Zeng, Yang Yu, and Yinghui Xu. Reinforcement learning to rank in E-commerce
search engine: Formalization, analysis, and application. In Proceedings of the ACM SIGKDD International
Conference on Knowledge Discovery and Data Mining , 2018.
[275] Jun Xu, Zeng Wei, Long Xia, Yanyan Lan, Dawei Yin, Xueqi Cheng, and Ji Rong Wen. Reinforcement Learning
to Rank with Pairwise Policy Gradient. In SIGIR 2020 - Proceedings of the 43rd International ACM SIGIR
Conference on Research and Development in Information Retrieval , 2020.
[276] Richard S. Sutton and Andrew G. Barto. Reinforcement Learning, Second Edition: An Introduction - Complete
Draft. The MIT Press , 2018.
[277] Marco Wiering and Martijn van Otterlo. Reinforcement Learning: State-of-the-Art , volume 12. Springer Berlin,
Heidelberg, 2012.
[278] Peng Jiang, Jiafeng Ma, and Jianming Zhang. Deep Reinforcement Learning based Recommender System with
State Representation. In Proceedings - 2021 IEEE International Conference on Big Data, Big Data 2021 , 2021.
[279] Peng Jiang, Jiafeng Ma, and Jianming Zhang. Deep Reinforcement Learning based Recommender System with
State Representation. In Proceedings - 2021 IEEE International Conference on Big Data, Big Data 2021 , 2021.
[280] Richard Liaw, Paige Bailey, Ying Li, Maria Dimakopoulou, and Yves Raimond. REVEAL 2022: Reinforcement
Learning-Based Recommender Systems at Scale. RecSys 2022 - Proceedings of the 16th ACM Conference on
Recommender Systems , 2022:684–685, 9 2022.
[281] Lu Wang, Xiaofeng He, Wei Zhang, and Hongyuan Zha. Supervised reinforcement learning with recurrent neural
network for dynamic treatment recommendation. In Proceedings of the ACM SIGKDD International Conference
on Knowledge Discovery and Data Mining , 2018.
[282] Wayne Xin Zhao, Kun Zhou, Junyi Li, Tianyi Tang, Xiaolei Wang, Yupeng Hou, Yingqian Min, Beichen Zhang,
Junjie Zhang, Zican Dong, Yifan Du, Chen Yang, Yushuo Chen, Zhipeng Chen, Jinhao Jiang, Ruiyang Ren,
Yifan Li, Xinyu Tang, Zikang Liu, Peiyu Liu, Jian-Yun Nie, and Ji-Rong Wen. A Survey of Large Language
Models. arXiv , 3 2023.
[283] Wang-cheng Kang, Wang-Cheng Kang, Jianmo Ni, Nikhil Mehta, Maheswaran Sathiamoorthy, Lichan Hong,
Ed Chi, and Derek Zhiyuan Cheng. Do LLMs Understand User Preferences? Evaluating LLMs On User Rating
Prediction. Papers with Code , 5 2023.
[284] Arpita Vats, Vinija Jain, Rahul Raja, and Aman Chadha. Exploring the Impact of Large Language Models on
Recommender Systems: An Extensive Review. 2 2024.
[285] Ross Taylor, Marcin Kardas, Guillem Cucurull, Thomas Scialom, Anthony Hartshorn, Elvis Saravia, Andrew
Poulton, Viktor Kerkez, and Robert Stojnic. Galactica: A Large Language Model for Science. 11 2022.
[286] Jianchao Ji, Zelong Li, Shuyuan Xu, Wenyue Hua, Yingqiang Ge, Juntao Tan, and Yongfeng Zhang. GenRec:
Large Language Model for Generative Recommendation. arXiv , 7 2023.
[287] Scott Sanner, Krisztian Balog, Filip Radlinski, Ben Wedin, and Lucas Dixon. Large Language Models are
Competitive Near Cold-start Recommenders for Language- and Item-based Preferences. In Proceedings of the
17th ACM Conference on Recommender Systems, RecSys 2023 , 2023.
[288] Yupeng Hou, Junjie Zhang, Zihan Lin, Hongyu Lu, Ruobing Xie, Julian McAuley, and Wayne Xin Zhao. Large
Language Models are Zero-Shot Rankers for Recommender Systems. pages 364–381, 5 2023.
[289] Zhankui He, Zhouhang Xie, Rahul Jha, Harald Steck, Dawen Liang, Yesu Feng, Bodhisattwa Prasad Majumder,
Nathan Kallus, and Julian McAuley. Large Language Models as Zero-Shot Conversational Recommenders.
International Conference on Information and Knowledge Management, Proceedings , pages 720–730, 10 2023.
[290] Junling Liu, Chao Liu, Peilin Zhou, Qichen Ye, Dading Chong, Kang Zhou, Yueqi Xie, Yuwei Cao, Shoujin
Wang, Chenyu You, and Philip S. Yu. LLMRec: Benchmarking Large Language Models on Recommendation
Task. 8 2023.
[291] Karthik Valmeekam, Matthew Marquez, Sarath Sreedharan, and Subbarao Kambhampati. On the Planning
Abilities of Large Language Models : A Critical Investigation. 5 2023.
[292] Fan Yang, Amazon AI Alexa Seattle, Usa Zheng Chen, Ziyan Jiang, Usa Eunah Cho, Usa Xiaojiang Huang,
Usa Yanbin Lu, Zheng Chen, Eunah Cho, and Xiaojiang Huang. PALR: Personalization Aware LLMs for
Recommendation. arXiv , 5 2023.
53
Page 54:
arXiv Template A P REPRINT
[293] Yancheng Wang, Ziyan Jiang, Zheng Chen, Fan Yang, Yingxue Zhou, Eunah Cho, Xing Fan, Xiaojiang Huang,
Yanbin Lu, and Yingzhen Yang. RecMind: Large Language Model Powered Agent For Recommendation. arXiv ,
8 2024.
[294] Xiaolei Wang, Xinyu Tang, Wayne Xin Zhao, Jingyuan Wang, and Ji-Rong Wen. Rethinking the Evaluation for
Conversational Recommendation in the Era of Large Language Models. arXiv , 5 2023.
[295] Keqin Bao, Jizhi Zhang, Yang Zhang, Wenjie Wang, Fuli Feng, and Xiangnan He. TALLRec: An Effective and
Efficient Tuning Framework to Align Large Language Model with Recommendation. In Proceedings of the 17th
ACM Conference on Recommender Systems, RecSys 2023 , 2023.
[296] Viomesh Kumar Singh, Sangeeta Sabharwal, and Goldie Gabrani. Comprehensive Analysis of Multimodal
Recommender Systems. In Data Intelligence and Cognitive Informatics , chapter Chapter 76, pages 887–901.
Springer, Singapore, 2021.
[297] Xin Zhou. MMRec: Simplifying Multimodal Recommendation. ACM Multimedia Asia Workshops , 8 2023.
[298] Wei Wei, Chao Huang, Lianghao Xia, and Chuxu Zhang. Multi-Modal Self-Supervised Learning for Recom-
mendation. In ACM Web Conference 2023 - Proceedings of the World Wide Web Conference, WWW 2023 ,
2023.
[299] Xu Chen, Hanxiong Chen, Hongteng Xu, Yongfeng Zhang, Yixin Cao, Zheng Qin, and Hongyuan Zha. Personal-
ized fashion recommendation with visual explanations based on multimodal attention network: Towards visually
explainable recommendation. In SIGIR 2019 - Proceedings of the 42nd International ACM SIGIR Conference on
Research and Development in Information Retrieval , 2019.
[300] Wei Wei, Jiabin Tang, Yangqin Jiang, Lianghao Xia, Chao Huang, and Chao 2024 Huang. PromptMM:
Multi-Modal Knowledge Distillation for Recommendation with Prompt-Tuning. arXiv , 2 2024.
[301] Xin Zhou, Hongyu Zhou, Yong Liu, Zhiwei Zeng, Chunyan Miao, Pengwei Wang, Yuan You, and Feijun
Jiang. Bootstrap Latent Representations for Multi-modal Recommendation. In ACM Web Conference 2023 -
Proceedings of the World Wide Web Conference, WWW 2023 , 2023.
[302] Risto Vuorio, Shao-Hua Sun, Hexiang Hu, Joseph J Lim, and SK T-Brain. Toward Multimodal Model-Agnostic
Meta-Learning. Papers with Code , 12 2018.
[303] Pushpendra Kumar and Ramjeevan Singh Thakur. Recommendation system techniques and related issues: a
survey. International Journal of Information Technology , 10:495–501, 2018.
[304] Xiangyu Zhao, Changsheng Gu, Haoshenglun Zhang, Xiwang Yang, Xiaobing Liu, Jiliang Tang, and Hui Liu.
Dear: Deep reinforcement learning for online advertising impression in recommender systems. In Proceedings
of the AAAI conference on artificial intelligence , volume 35, pages 750–758, 2021.
[305] Tarana Singh, Anand Nayyar, and Arun Solanki. Multilingual opinion mining movie recommendation system
using rnn. In Proceedings of first international conference on computing, communications, and cyber-security
(IC4S 2019) , pages 589–605. Springer, 2020.
[306] Ruslan Salakhutdinov and Andriy Mnih. Bayesian probabilistic matrix factorization using markov chain Monte
Carlo. In Proceedings of the 25th International Conference on Machine Learning , 2008.
[307] Francisco J. Peña, DIarmuid O’Reilly-Morgan, Elias Z. Tragos, Neil Hurley, Erika Duriakova, Barry Smyth, and
Aonghus Lawlor. Combining Rating and Review Data by Initializing Latent Factor Models with Topic Models
for Top-N Recommendation. In RecSys 2020 - 14th ACM Conference on Recommender Systems , 2020.
[308] Ruining He, Wang Cheng Kang, and Julian McAuley. Translation-based recommendation. RecSys 2017 -
Proceedings of the 11th ACM Conference on Recommender Systems , pages 161–169, 8 2017.
[309] Antoine Bordes, Nicolas Usunier, Alberto Garcia-Durán, Jason Weston, and Oksana Yakhnenko. Translating
embeddings for modeling multi-relational data. In Advances in Neural Information Processing Systems , 2013.
[310] Aditya Grover and Jure Leskovec. Node2vec: Scalable feature learning for networks. In Proceedings of the ACM
SIGKDD International Conference on Knowledge Discovery and Data Mining , volume 13-17-August-2016,
2016.
[311] Deqing Yang, Zikai Guo, Ziyi Wang, Juyang Jiang, Yanghua Xiao, and Wei Wang. A Knowledge-Enhanced
Deep Recommendation Framework Incorporating GAN-Based Models. In Proceedings - IEEE International
Conference on Data Mining, ICDM , volume 2018-November, 2018.
[312] Hongwei Wang, Fuzheng Zhang, Min Hou, Xing Xie, Minyi Guo, and Qi Liu. SHINE: Signed heterogeneous
information network embedding for sentiment link prediction. In WSDM 2018 - Proceedings of the 11th ACM
International Conference on Web Search and Data Mining , volume 2018-Febuary, 2018.
54
Page 55:
arXiv Template A P REPRINT
[313] Binbin Hu, Chuan Shi, Wayne Xin Zhao, and Philip S. Yu. Leveraging meta-path based context for top-
n recommendation with a neural co-attention model. In Proceedings of the ACM SIGKDD International
Conference on Knowledge Discovery and Data Mining , 2018.
[314] Xiao Yu, Xiang Ren, Yizhou Sun, Quanquan Gu, Bradley Sturt, Urvashi Khandelwal, Brandon Norick, and
Jiawei Han. Personalized entity recommendation: A heterogeneous information network approach. In WSDM
2014 - Proceedings of the 7th ACM International Conference on Web Search and Data Mining , 2014.
[315] Chen Luo, Wei Pang, Zhe Wang, and Chenghua Lin. Hete-CF: Social-Based Collaborative Filtering Recommen-
dation Using Heterogeneous Relations. In Proceedings - IEEE International Conference on Data Mining, ICDM ,
volume 2015-January, 2014.
[316] Chuan Shi, Zhiqiang Zhang, Ping Luo, Philip S. Yu, Yading Yue, and Bin Wu. Semantic path based personalized
recommendation on weighted heterogeneous information networks. In International Conference on Information
and Knowledge Management, Proceedings , volume 19-23-Oct-2015, 2015.
[317] Xin Xin, Xiangnan He, Yongfeng Zhang, Yongdong Zhang, and Joemon Jose. Relational collaborative filtering:
Modeling multiple item relations for recommendation. In SIGIR 2019 - Proceedings of the 42nd International
ACM SIGIR Conference on Research and Development in Information Retrieval , 2019.
[318] Xiaoli Tang, Tengyun Wang, Haizhi Yang, and Hengjie Song. Akupm: Attention-enhanced knowledge-aware
user preference model for recommendation. In Proceedings of the ACM SIGKDD International Conference on
Knowledge Discovery and Data Mining , 2019.
[319] Yanru Qu, Ting Bai, Weinan Zhang, Jianyun Nie, and Jian Tang. An End-to-end neighborhood-based interaction
model for knowledge-enhanced recommendation. In Proceedings of the ACM SIGKDD International Conference
on Knowledge Discovery and Data Mining , 2019.
[320] Kun Zhou, Wayne Xin Zhao, Shuqing Bian, Yuanhang Zhou, Ji-Rong Wen, and Jingsong Yu. Improving
Conversational Recommender Systems via Knowledge Graph based Semantic Fusion. In Proceedings of the 26th
ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , KDD ’20, pages 1006–1014,
New York, NY , USA, 2020. Association for Computing Machinery.
[321] Tao Qi, Fangzhao Wu, Chuhan Wu, and Yongfeng Huang. Personalized News Recommendation with Knowledge-
aware Interactive Matching. In SIGIR 2021 - Proceedings of the 44th International ACM SIGIR Conference on
Research and Development in Information Retrieval , 2021.
[322] Armin Toroghi, Griffin Floto, Zhenwei Tang, and Scott Sanner. Bayesian Knowledge-driven Critiquing with In-
direct Evidence. In Proceedings of the 46th International ACM SIGIR Conference on Research and Development
in Information Retrieval , SIGIR ’23, pages 1838–1842, New York, NY , USA, 2023. Association for Computing
Machinery.
[323] Yangqin Jiang, Yuhao Yang, Lianghao Xia, and Chao Huang. DiffKG: Knowledge Graph Diffusion Model for
Recommendation. In Proceedings of the 17th ACM International Conference on Web Search and Data Mining ,
WSDM ’24, pages 313–321, New York, NY , USA, 2024. Association for Computing Machinery.
[324] Mahesan Niranjan. On-Line Q-Learning Using Connectionist Systems. 1994.
[325] Tariq Mahmood, Ghulam Mujtaba, and Adriano Venturini. Dynamic personalization in conversational recom-
mender systems. Information Systems and e-Business Management , 12(2), 2014.
[326] Cameron B. Browne, Edward Powley, Daniel Whitehouse, Simon M. Lucas, Peter I. Cowling, Philipp Rohlfsha-
gen, Stephen Tavener, Diego Perez, Spyridon Samothrakis, and Simon Colton. A survey of Monte Carlo tree
search methods, 2012.
[327] Chengrun Qiu, Yang Hu, Yan Chen, and Bing Zeng. Deep Deterministic Policy Gradient (DDPG)-Based Energy
Harvesting Wireless Communications. IEEE Internet of Things Journal , 6(5), 2019.
[328] Tuomas Haarnoja, Aurick Zhou, Kristian Hartikainen, George Tucker, Sehoon Ha, Jie Tan, Vikash Kumar, Henry
Zhu, Abhishek Gupta, Pieter Abbeel, and Sergey Levine. Soft Actor-Critic Algorithms and Applications. Paper
with Code , 2018.
[329] Feiyang Pan, Qingpeng Cai, Pingzhong Tang, Fuzhen Zhuang, and Qing He. Policy gradients for contextual
recommendations. In The Web Conference 2019 - Proceedings of the World Wide Web Conference, WWW 2019 ,
2019.
[330] Don X. Sun and Frederick Jelinek. Statistical Methods for Speech Recognition. Journal of the American
Statistical Association , 94(446), 1999.
[331] Ronald Rosenfeld. Two decdes of statistical language modeling where do we go form here? Where do we go
from here? Proceedings of the IEEE , 88(8):1270–1275, 2000.
55
Page 56:
arXiv Template A P REPRINT
[332] Yoshua Bengio, Réjean Ducharme, Pascal Vincent, and Christian Jauvin. A Neural Probabilistic Language
Model. In Journal of Machine Learning Research , volume 3, 2003.
[333] Tomaš Mikolov, Martin Karafiát, Lukaš Burget, Cernocky Jan, and Sanjeev Khudanpur. Recurrent neural
network based language model. In Proceedings of the 11th Annual Conference of the International Speech
Communication Association, INTERSPEECH 2010 , 2010.
[334] Stefan Kombrink, Tomáš Mikolov, Martin Karafiát, and Lukáš Burget. Recurrent neural network based language
modeling in meeting recognition. In Proceedings of the Annual Conference of the International Speech
Communication Association, INTERSPEECH , 2011.
[335] Jacob Devlin, Ming Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep bidirectional
transformers for language understanding. In NAACL HLT 2019 - 2019 Conference of the North American
Chapter of the Association for Computational Linguistics: Human Language Technologies - Proceedings of the
Conference , volume 1, 2019.
[336] Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Yunxuan Li, Xuezhi Wang,
Mostafa Dehghani, Siddhartha Brahma, Albert Webson, Shixiang Shane Gu, Zhuyun Dai, Mirac Suzgun, Xinyun
Chen, Aakanksha Chowdhery, Alex Castro-Ros, Marie Pellat, Kevin Robinson, Dasha Valter, Sharan Narang,
Gaurav Mishra, Adams Yu, Vincent Zhao, Yanping Huang, Andrew Dai, Hongkun Yu, Slav Petrov, Ed H. Chi,
Jeff Dean, Jacob Devlin, Adam Roberts, Denny Zhou, Quoc V . Le, and Jason Wei. Scaling Instruction-Finetuned
Language Models. 10 2022.
[337] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser,
and Illia Polosukhin. Attention is all you need. In Advances in neural information processing systems , pages
5998–6008, 2017.
[338] Md Tahmid Rahman Laskar, M Saiful Bari, Mizanur Rahman, Md Amran Hossen Bhuiyan, Shafiq Joty, and
Jimmy Xiangji Huang. A systematic study and comprehensive evaluation of chatgpt on benchmark datasets.
arXiv preprint arXiv:2305.18486 , 2023.
[339] Aashiq Muhamed, Iman Keivanloo, Sujan Perera, James Mracek, Yi Xu, Qingjun Cui, Santosh Rajagopalan,
Belinda Zeng, and Trishul Chilimbi. CTR-BERT: Cost-effective knowledge distillation for billion-parameter
teacher models. arXiv , 2021.
[340] Bowen Yang, Cong Han, Yu Li, Lei Zuo, and Zhou Yu. Improving Conversational Recommendation Systems’
Quality with Context-Aware Item Meta-Information. In Findings of the Association for Computational Linguistics:
NAACL 2022 - Findings , 2022.
[341] Zhaopeng Qiu, Xian Wu, Jingyue Gao, and Wei Fan. U-BERT: Pre-training User Representations for Improved
Recommendation. In 35th AAAI Conference on Artificial Intelligence, AAAI 2021 , volume 5B, 2021.
[342] Chuhan Wu, Fangzhao Wu, Tao Qi, and Yongfeng Huang. UserBERT: Pre-training User Model with Contrastive
Self-supervision. In SIGIR 2022 - Proceedings of the 45th International ACM SIGIR Conference on Research
and Development in Information Retrieval , 2022.
[343] Yingrui Yang, Yifan Qiao, Jinjin Shao, Xifeng Yan, and Tao Yang. Lightweight composite re-ranking for efficient
keyword search with BERT. In WSDM 2022 - Proceedings of the 15th ACM International Conference on Web
Search and Data Mining , 2022.
[344] Song Zhang, Nan Zheng, and Danli Wang. GBERT: Pre-training User representations for Ephemeral Group
Recommendation. In International Conference on Information and Knowledge Management, Proceedings , 2022.
[345] Shijie Geng, Shuchang Liu, Zuohui Fu, Yingqiang Ge, and Yongfeng Zhang. Recommendation as Language
Processing (RLP): A Unified Pretrain, Personalized Prompt and Predict Paradigm (P5). In RecSys 2022 -
Proceedings of the 16th ACM Conference on Recommender Systems , 2022.
[346] Zizhuo Zhang and Bang Wang. Prompt Learning for News Recommendation. In SIGIR 2023 - Proceedings of
the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval , 2023.
[347] Xuyang Wu, Alessandro Magnani, Suthee Chaidaroon, Ajit Puthenputhussery, Ciya Liao, and Yi Fang. A
Multi-task Learning Framework for Product Ranking with BERT. In WWW 2022 - Proceedings of the ACM Web
Conference 2022 , 2022.
[348] Sahil Verma, Ashudeep Singh, Varich Boonsanong, John P Dickerson, and Chirag Shah. RecRec: Algorithmic
Recourse for Recommender Systems. In CIKM ’23: Proceedings of the 32nd ACM International Conference on
Information and Knowledge Management , number 23, 2023.
[349] Ruining He and Julian McAuley. VBPR: Visual Bayesian personalized ranking from implicit feedback. In 30th
AAAI Conference on Artificial Intelligence, AAAI 2016 , 2016.
56
Page 57:
arXiv Template A P REPRINT
[350] Jingyuan Chen, Hanwang Zhang, Xiangnan He, Liqiang Nie, Wei Liu, and Tat Seng Chua. Attentive collaborative
filtering: Multimedia recommendation with item-And component-level attention. In SIGIR 2017 - Proceedings
of the 40th International ACM SIGIR Conference on Research and Development in Information Retrieval , 2017.
[351] Guangneng Hu, Yu Zhang, and Qiang Yang. Conet: Collaborative cross networks for cross-domain recommen-
dation. In International Conference on Information and Knowledge Management, Proceedings , 2018.
[352] Gediminas Adomavicius, Konstantin Bauman, Alexander Tuzhilin, and Moshe Unger. Context-Aware Rec-
ommender Systems: From Foundations to Recent Developments. In Recommender Systems Handbook: Third
Edition , chapter Chapter 5, pages 211–250. Springer, New York, NY , 2022.
[353] Steffen Rendle. Factorization machines. Proceedings - IEEE International Conference on Data Mining, ICDM ,
pages 995–1000, 2010.
[354] Yuchin Juan, Yong Zhuang, Wei Sheng Chin, and Chih Jen Lin. Field-aware factorization machines for CTR
prediction. In RecSys 2016 - Proceedings of the 10th ACM Conference on Recommender Systems , 2016.
[355] Mingmin Jin, Xin Luo, Huiling Zhu, Hankz Hankui Zhuo, and China Jinmm. Combining Deep Learning and
Topic Modeling for Review Understanding in Context-Aware Recommendation Guangdong Key Laboratory of
Big Data Analysis and Processing. In Proceedings of the 2018 Conference of the North American Chapter of the
Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers). , 2018.
[356] Jun Xiao, Hao Ye, Xiangnan He, Hanwang Zhang, Fei Wu, and Tat Seng Chua. Attentional factorization
machines: Learning the weight of feature interactions via attention networks. In IJCAI International Joint
Conference on Artificial Intelligence , volume 0, 2017.
[357] Angela Carrera-Rivera, Felix Larrinaga, and Ganix Lasa. Context-awareness for the design of Smart-product
service systems: Literature review, 2022.
[358] Emrul Hasan, Mizanur Rahman, Chen Ding, Jimmy Xiangji Huang, and Shaina Raza. based recommender
systems: A survey of approaches, challenges and future perspectives. arXiv preprint arXiv:2405.05562 , 2024.
[359] Julian McAuley and Jure Leskovec. Hidden factors and hidden topics: Understanding rating dimensions with
review text. In RecSys 2013 - Proceedings of the 7th ACM Conference on Recommender Systems , 2013.
[360] Yunzhi Tan, Min Zhang, Yiqun Liu, and Shaoping Ma. Rating-boosted latent topics: Understanding users
and items with ratings and reviews. In IJCAI International Joint Conference on Artificial Intelligence , volume
2016-January, 2016.
[361] Donghyun Kim, Chanyoung Park, Jinoh Oh, Sungyoung Lee, and Hwanjo Yu. Convolutional matrix factorization
for document context-aware recommendation. RecSys 2016 - Proceedings of the 10th ACM Conference on
Recommender Systems , pages 233–240, 9 2016.
[362] Chuhan Wu, Fangzhao Wu, Tao Qi, and Yongfeng Huang. SentiRec: Sentiment Diversity-aware Neural News
Recommendation. pages 44–53.
[363] Jing Liu, Jinbao Song, Chen Li, Xiaoya Zhu, and Ruyi Deng. A Hybrid News Recommendation Algorithm
Based on K-means Clustering and Collaborative Filtering. In Journal of Physics: Conference Series , volume
1881, 2021.
[364] Zhiyong Cheng, Ying Ding, Xiangnan He, Lei Zhu, Xuemeng Song, and Mohan Kankanhalli. A3NCF: An
Adaptive Aspect Attention Model for Rating Prediction. 2018.
[365] Xinyu Guan, Zhiyong Cheng, Xiangnan He, Yongfeng Zhang, Zhibo Zhu, Qinke Peng, and Tat-Seng Chua.
Attentive Aspect Modeling for Review-aware Recommendation. ACM Transactions on Information Systems ,
37(3):28, 11 2018.
[366] Songyin Luo, Xiangkui Lu, Jun Wu, and Jianbo Yuan. Aware neural recommendation with cross-modality mutual
attention. In Proceedings of the 30th ACM International Conference on Information & Knowledge Management ,
pages 3293–3297, 2021.
[367] Sungyong Seo, Jing Huang, Hao Yang, and Yan Liu. Interpretable convolutional neural networks with dual local
and global attention for review rating prediction. In RecSys 2017 - Proceedings of the 11th ACM Conference on
Recommender Systems , 2017.
[368] Chong Chen, Min Zhang, Yiqun Liu, and Shaoping Ma. Neural attentional rating regression with review-level
explanations. In The Web Conference 2018 - Proceedings of the World Wide Web Conference, WWW 2018 , 2018.
[369] Hongtao Liu, Fangzhao Wu, Wenjun Wang, Xianchen Wang, Pengfei Jiao, Chuhan Wu, and Xing Xie. NRPA:
Neural Recommendation with Personalized Attention. SIGIR 2019 - Proceedings of the 42nd International ACM
SIGIR Conference on Research and Development in Information Retrieval , 19:1233–1236, 5 2019.
57
Page 58:
arXiv Template A P REPRINT
[370] Yichao Lu, Ruihai Dong, and Barry Smyth. Coevolutionary recommendation model: Mutual learning between
ratings and reviews. In Proceedings of the 2018 World Wide Web Conference , pages 773–782, 2018.
[371] Xin Dong, Jingchao Ni, Wei Cheng, Zhengzhang Chen, Bo Zong, Dongjin Song, Yanchi Liu, Haifeng Chen, and
Gerard De Melo. Asymmetrical hierarchical networks with attentive interactions for interpretable review-based
recommendation. In Proceedings of the AAAI conference on artificial intelligence , volume 34, pages 7667–7674,
2020.
[372] Ke Wang, Yanmin Zhu, Haobing Liu, Tianzi Zang, and Chunyang Wang. Learning Aspect-Aware High-Order
Representations from Ratings and Reviews for Recommendation. ACM Transactions on Knowledge Discovery
from Data , 17(1), 2023.
[373] Rose Catherine and William Cohen. TransNets: Learning to transform for recommendation. In RecSys 2017 -
Proceedings of the 11th ACM Conference on Recommender Systems , 2017.
[374] Donghua Liu, Jing Li, Bo Du, Jun Chang, and Rong Gao. DAML: Dual attention mutual learning between
ratings and reviews for item recommendation. Proceedings of the ACM SIGKDD International Conference on
Knowledge Discovery and Data Mining , pages 344–352, 7 2019.
[375] Peng Liu, Lemei Zhang, and Jon Atle Gulla. Multilingual Review-Aware Deep Recommender System via
Aspect-based Sentiment Analysis. 2020.
[376] Jin Yao Chin, Shafiq Joty, Kaiqi Zhao, and Gao Cong. ANR: Aspect-based Neural Recommender. In International
Conference on Information and Knowledge Management, Proceedings , 2018.
[377] Konstantin Bauman, Bing Liu, and Alexander Tuzhilin. Aspect based recommendations: Recommending items
with the most valuable aspects based on user reviews. In Proceedings of the ACM SIGKDD International
Conference on Knowledge Discovery and Data Mining , volume Part F129685, 2017.
[378] Cataldo Musto, Marco De Gemmis, Giovanni Semeraro, and Pasquale Lops. A multi-criteria recommender
system exploiting aspect-based sentiment analysis of users’ reviews. In RecSys 2017 - Proceedings of the 11th
ACM Conference on Recommender Systems , 2017.
[379] Liliya V olkova, Elena Yagunova, Ekaterina Pronoza, Alexandra Maslennikova, Danil Bliznuk, Margarita
Tokareva, and Ali Abdullaev. Recommender System for Tourist Itineraries Based on Aspects Extraction
from Reviews Corpora. Polibits , 57, 2018.
[380] Aminu Da’u, Naomie Salim, Idris Rabiu, and Akram Osman. Weighted aspect-based opinion mining using deep
learning for recommender system. Expert Systems with Applications , 140, 2020.
[381] Aminu Da’u, Naomie Salim, Idris Rabiu, and Akram Osman. Recommendation system exploiting aspect-based
opinion mining with deep learning method. Information Sciences , 512, 2020.
[382] Ahlem Drif, Sami Guembour, and Hocine Cherifi. A Sentiment Enhanced Deep Collaborative Filtering Recom-
mender System. In Studies in Computational Intelligence , volume 944, 2021.
[383] Avinash Bhojwani, Vanshika Jolly, Saksham Goel, and M. Anand Kumar. Aspect Based Neural Recommender
Using Adaptive Prediction. In 2023 IEEE International Students’ Conference on Electrical, Electronics and
Computer Science, SCEECS 2023 , 2023.
[384] Hanning Yuan, Zhengyu Chen, Jingting Yang, Shuliang Wang, Jing Geng, and Chuwen Ke. A Hybrid Aspect
Based Latent Factor Model for Recommendation. Chinese Journal of Electronics , 29(3), 2020.
[385] Kexin Yin, Xiao Fang, Bintong Chen, and Olivia R.Liu Sheng. Diversity Preference-Aware Link Recommenda-
tion for Online Social Networks. Information Systems Research , 34(4), 2023.
[386] Payal Mehra. Unexpected surprise: Emotion analysis and aspect based sentiment analysis (ABSA) of user
generated comments to study behavioral intentions of tourists. Tourism Management Perspectives , 45:101063, 1
2023.
[387] Hengyun Li, Bruce X.B. Yu, Gang Li, and Huicai Gao. Restaurant survival prediction using customer-generated
content: An aspect-based sentiment analysis of online reviews. Tourism Management , 96:104707, 6 2023.
[388] Shoujin Wang, Xiuzhen Zhang, Yan Wang, and Francesco Ricci. Trustworthy recommender systems. ACM
Transactions on Intelligent Systems and Technology , 2022.
[389] Yongfeng Zhang, Guokun Lai, Min Zhang, Yi Zhang, Yiqun Liu, and Shaoping Ma. Explicit Factor Models for
explainable recommendation based on phrase-level sentiment analysis. In SIGIR 2014 - Proceedings of the 37th
International ACM SIGIR Conference on Research and Development in Information Retrieval , 2014.
[390] Xiangnan He, Tao Chen, Min Yen Kan, and Xiao Chen. TriRank: Review-aware explainable recommendation
by modeling aspects. In International Conference on Information and Knowledge Management, Proceedings ,
volume 19-23-Oct-2015, 2015.
58
Page 59:
arXiv Template A P REPRINT
[391] Xiang Wang, Xiangnan He, Fuli Feng, Liqiang Nie, and Tat Seng Chua. TEM: Tree-enhanced embedding model
for explainable recommendation. In The Web Conference 2018 - Proceedings of the World Wide Web Conference,
WWW 2018 , 2018.
[392] Qingyao Ai, Vahid Azizi, Xu Chen, and Yongfeng Zhang. Learning heterogeneous knowledge base embeddings
for explainable recommendation. Algorithms , 11(9), 2018.
[393] Zhiyong Cheng, Xiaojun Chang, Lei Zhu, Rose C. Kanjirathinkal, and Mohan Kankanhalli. MMalfM: Explain-
able recommendation by leveraging reviews and images. ACM Transactions on Information Systems , 37(2),
2019.
[394] Krisztian Balog, Filip Radlinski, and Shushan Arakelyan. Transparent, scrutable and explainable user models for
personalized recommendation. In SIGIR 2019 - Proceedings of the 42nd International ACM SIGIR Conference
on Research and Development in Information Retrieval , 2019.
[395] Lei Li, Yongfeng Zhang, and Li Chen. Personalized transformer for explainable recommendation. In ACL-
IJCNLP 2021 - 59th Annual Meeting of the Association for Computational Linguistics and the 11th International
Joint Conference on Natural Language Processing, Proceedings of the Conference , 2021.
[396] Juntao Tan, Shuyuan Xu, Yingqiang Ge, Yunqi Li, Xu Chen, and Yongfeng Zhang. Counterfactual Explainable
Recommendation. In International Conference on Information and Knowledge Management, Proceedings , 2021.
[397] Yingqiang Ge, Juntao Tan, Yan Zhu, Yinglong Xia, Jiebo Luo, Shuchang Liu, Zuohui Fu, Shijie Geng, Zelong
Li, and Yongfeng Zhang. Explainable Fairness in Recommendation. In SIGIR 2022 - Proceedings of the 45th
International ACM SIGIR Conference on Research and Development in Information Retrieval , 2022.
[398] Lei Li, Yongfeng Zhang, and Li Chen. Personalized prompt learning for explainable recommendation. ACM
Transactions on Information Systems , 41(4):1–26, 2023.
[399] Tianjun Wei, Tommy WS Chow, Jianghong Ma, and Mingbo Zhao. Expgcn: Review-aware graph convolution
network for explainable recommendation. Neural Networks , 157:202–215, 2023.
[400] Nasim Sonboli, Jessie J. Smith, Florencia Cabral Berenfus, Robin Burke, and Casey Fiesler. Fairness and
transparency in recommendation: The users’ perspective. In UMAP 2021 - Proceedings of the 29th ACM
Conference on User Modeling, Adaptation and Personalization , 2021.
[401] Yifan Wang, Weizhi Ma, Min Zhang, Yiqun Liu, and Shaoping Ma. A Survey on the Fairness of Recommender
Systems. ACM Transactions on Information Systems , 41(3), 2023.
[402] Bashir Rastegarpanah, Krishna P. Gummadi, and Mark Crovella. Fighting fire with fire: Using antidote data to
improve polarization and fairness of recommender systems. In WSDM 2019 - Proceedings of the 12th ACM
International Conference on Web Search and Data Mining , 2019.
[403] Sirui Yao and Bert Huang. Beyond parity: Fairness objectives for collaborative filtering. In Advances in Neural
Information Processing Systems , volume 2017-December, 2017.
[404] Toshihiro Kamishima, Shotaro Akaho, Hideki Asoh, and Jun Sakuma. Recommendation Independence. In
Proceedings of Machine Learning Research , volume 81, 2018.
[405] Ziwei Zhu, Xia Hu, and James Caverlee. Fairness-aware tensor-based recommendation. In International
Conference on Information and Knowledge Management, Proceedings , 2018.
[406] Alex Beutel, Jilin Chen, Tulsee Doshi, Hai Qian, Li Wei, Yi Wu, Lukasz Heldt, Zhe Zhao, Lichan Hong, Ed H.
Chi, and Cristos Goodrow. Fairness in recommendation ranking through pairwise comparisons. In Proceedings
of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , 2019.
[407] Mengting Wan, Jianmo Ni, Rishabh Misra, and Julian McAuley. Addressing marketing bias in product rec-
ommendations. In WSDM 2020 - Proceedings of the 13th International Conference on Web Search and Data
Mining , 2020.
[408] Shaina Raza and Chen Ding. A Regularized Model to Trade-off between Accuracy and Diversity in a News
Recommender System. In Proceedings - 2020 IEEE International Conference on Big Data, Big Data 2020 ,
2020.
[409] Shaina Raza. Bias Reduction News Recommendation System. Digital 2024, Vol. 4, Pages 92-103 , 4(1):92–103,
12 2023.
[410] Ziwei Zhu, Jianling Wang, and James Caverlee. Fairness-aware personalized ranking recommendation via
adversarial learning. arXiv preprint arXiv:2103.07849 , 2021.
[411] Yunqi Li, Hanxiong Chen, Shuyuan Xu, Yingqiang Ge, and Yongfeng Zhang. Towards Personalized Fairness
based on Causal Notion. In SIGIR 2021 - Proceedings of the 44th International ACM SIGIR Conference on
Research and Development in Information Retrieval , 2021.
59
Page 60:
arXiv Template A P REPRINT
[412] Yingqiang Ge, Shuchang Liu, Ruoyuan Gao, Yikun Xian, Yunqi Li, Xiangyu Zhao, Changhua Pei, Fei Sun,
Junfeng Ge, Wenwu Ou, and Yongfeng Zhang. Towards Long-term Fairness in Recommendation. In WSDM
2021 - Proceedings of the 14th ACM International Conference on Web Search and Data Mining , 2021.
[413] Rodrigo Borges and Kostas Stefanidis. Enhancing long term fairness in recommendations with variational
autoencoders. In 11th International Conference on Management of Digital EcoSystems, MEDES 2019 , 2019.
[414] Rashidul Islam, Kamrun Naher Keya, Ziqian Zeng, Shimei Pan, and James Foulds. Debiasing career recommen-
dations with neural fair collaborative filtering. In The Web Conference 2021 - Proceedings of the World Wide
Web Conference, WWW 2021 , 2021.
[415] Yangkun Li, Mohamed Laid Hedia, Weizhi Ma, Hongyu Lu, Min Zhang, Yiqun Liu, and Shaoping Ma.
Contextualized Fairness for Recommender Systems in Premium Scenarios. Big Data Research , 27, 2022.
[416] Meike Zehlike, Francesco Bonchi, Carlos Castillo, Sara Hajian, Mohamed Megahed, and Ricardo Baeza-Yates.
FAIR: A fair top-k ranking algorithm. In International Conference on Information and Knowledge Management,
Proceedings , volume Part F131841, 2017.
[417] Harald Steck. Calibrated recommendations. In RecSys 2018 - 12th ACM Conference on Recommender Systems ,
2018.
[418] Sahin Cem Geyik, Stuart Ambler, and Krishnaram Kenthapadi. Fairness-aware ranking in search & recommen-
dation systems with application to linkedin talent search. In Proceedings of the ACM SIGKDD International
Conference on Knowledge Discovery and Data Mining , 2019.
[419] Weiwen Liu, Jun Guo, Nasim Sonboli, Robin Burke, and Shengyu Zhang. Personalized fairness-aware re-ranking
for microlending. In RecSys 2019 - 13th ACM Conference on Recommender Systems , 2019.
[420] Xiao Lin, Min Zhang, Yongfeng Zhang, Zhaoquan Gu, Yiqun Liu, and Shaoping Ma. Fairness-aware group
recommendation with pareto-efficiency. In RecSys 2017 - Proceedings of the 11th ACM Conference on Recom-
mender Systems , 2017.
[421] Ziwei Zhu, Jingu Kim, Trung Nguyen, Aish Fenton, and James Caverlee. Fairness among New Items in Cold
Start Recommender Systems. In SIGIR 2021 - Proceedings of the 44th International ACM SIGIR Conference on
Research and Development in Information Retrieval , 2021.
[422] Di Jin, Luzhi Wang, He Zhang, Yizhen Zheng, Weiping Ding, Feng Xia, and Shirui Pan. A survey on fairness-
aware recommender systems. Information Fusion , 100:101906, 12 2023.
[423] Yao Wu, Jian Cao, and Guandong Xu. Fairness in Recommender Systems: Evaluation Approaches and Assurance
Strategies. ACM Transactions on Knowledge Discovery from Data , 18(1), 8 2023.
[424] Sriharsha Dara, C. Ravindranath Chowdary, and Chintoo Kumar. A survey on group recommender systems.
Journal of Intelligent Information Systems , 54(2), 2020.
[425] Xuan Bi, Annie Qu, Junhui Wang, and Xiaotong Shen. A Group-Specific Recommender System. Journal of the
American Statistical Association , 112(519), 2017.
[426] Nguyen Thanh Toan, Phan Thanh Cong, Nguyen Thanh Tam, Nguyen Quoc Viet Hung, and Bela Stantic.
Diversifying Group Recommendation. IEEE Access , 6, 2018.
[427] Nicola Capuano, Francisco Chiclana, Enrique Herrera-Viedma, Hamido Fujita, and Vincenzo Loia. Fuzzy Group
Decision Making for influence-aware recommendations. Computers in Human Behavior , 101, 2019.
[428] Zhiwei Guo, Wenru Zeng, Heng Wang, and Yu Shen. An Enhanced Group Recommender System by Exploiting
Preference Relation. IEEE Access , 7, 2019.
[429] Hongzhi Yin, Qinyong Wang, Kai Zheng, Zhixu Li, Jiali Yang, and Xiaofang Zhou. Social influence-based
group representation learning for group recommendation. In Proceedings - International Conference on Data
Engineering , volume 2019-April, 2019.
[430] Pablo Sánchez and Alejandro Bellogín. On the effects of aggregation strategies for different groups of users in
venue recommendation. Information Processing and Management , 58(5), 2021.
[431] Firat Ismailoglu. Aggregating user preferences in group recommender systems: A crowdsourcing approach.
Decision Support Systems , 152, 2022.
[432] Himan Abdollahpouri, Gediminas Adomavicius, Robin Burke, Ido Guy, Dietmar Jannach, Toshihiro Kamishima,
Jan Krasnodebski, and Luiz Pizzato. Multistakeholder recommendation: Survey and research directions. User
Modeling and User-Adapted Interaction , 30(1), 2020.
[433] Anitha Anandhan, Liyana Shuib, Maizatul Akmar Ismail, and Ghulam Mujtaba. Social Media Recommender
Systems: Review and Open Research Issues. IEEE Access , 6, 2018.
60
Page 61:
arXiv Template A P REPRINT
[434] Jyoti Shokeen and Chhavi Rana. A study on features of social recommender systems. Artificial Intelligence
Review , 53(2), 2020.
[435] Amazon’s Product Recommendation System In 2021: How Does The Algorithm Of The eCommerce Giant
Work? - Recostream.
[436] Yan Guo, Minxi Wang, and Xin Li. Application of an improved Apriori algorithm in a mobile e-commerce
recommendation system. Industrial Management and Data Systems , 117(2), 2017.
[437] Yulong Gu, Zhuoye Ding, Shuaiqiang Wang, Lixin Zou, Yiding Liu, and Dawei Yin. Deep Multifaceted
Transformers for Multi-objective Ranking in Large-Scale E-commerce Recommender Systems. In International
Conference on Information and Knowledge Management, Proceedings , 2020.
[438] Carlos A. Gomez-Uribe and Neil Hunt. The netflix recommender system: Algorithms, business value, and
innovation. ACM Transactions on Management Information Systems , 6(4), 2015.
[439] Harald Steck, Linas Baltrunas, Ehtsham Elahi, Dawen Liang, Yves Raimond, and Justin Basilico. Deep learning
for recommender systems: A Netflix case study. AI Magazine , 42(3), 2021.
[440] Chhavi Maheshwari. Music Recommendation on Spotify using Deep Learning. 12 2023.
[441] Kurt Jacobson, Vidhya Murali, Edward Newett, Brian Whitman, and Romain Yon. Music Personalization at
Spotify. In RecSys ’16: Proceedings of the 10th ACM Conference on Recommender Systems , pages 373–373.
Association for Computing Machinery (ACM), 9 2016.
[442] Germán Cheuque, José Guzmán, and Denis Parra. Recommender systems for online video game platforms: The
case of steam. In The Web Conference 2019 - Companion of the World Wide Web Conference, WWW 2019 , 2019.
[443] Hanna Schäfer, Santiago Hors-Fraile, Raghav Pavan Karumur, André Calero Valdez, Alan Said, Helma Torka-
maan, Tom Ulmer, and Christoph Trattner. Towards health (Aware) recommender systems. In ACM International
Conference Proceeding Series , volume Part F128634, 2017.
[444] Juan G.Diaz Ochoa, Orsolya Csiszár, and Thomas Schimper. Medical recommender systems based on continuous-
valued logic and multi-criteria decision operators, using interpretable neural networks. BMC Medical Informatics
and Decision Making , 21(1), 2021.
[445] Deepika Sharma, Gagangeet Singh Aujla, and Rohit Bajaj. Evolution from ancient medication to human-centered
Healthcare 4.0: A review on health care recommender systems. International Journal of Communication Systems ,
36(12), 2023.
[446] Uzair Aslam Bhatti, Mengxing Huang, Di Wu, Yu Zhang, Anum Mehmood, and Huirui Han. Recommendation
system using feature extraction and pattern recognition in clinical care systems. Enterprise Information Systems ,
13(3), 2019.
[447] André Calero Valdez, Martina Ziefle, Katrien Verbert, Alexander Felfernig, and Andreas Holzinger. Recom-
mender systems for health informatics: State-of-the-art and future perspectives. In Machine Learning for Health
Informatics , chapter Chapter 21, pages 391–414. Springer, Cham, 2016.
[448] Donghui Yang, Chao Huang, and Mingyang Wang. A social recommender system by combining social network
and sentiment similarity: A case study of healthcare. Journal of Information Science , 43(5), 2017.
[449] Anam Mustaqeem, Syed Muhammad Anwar, Abdul Rashid Khan, and Muhammad Majid. A statistical analysis
based recommender model for heart disease patients. International Journal of Medical Informatics , 108, 2017.
[450] Luciano Rodrigo Ferretto, Cristiano Roberto Cervi, and Ana Carolina Bertoletti De Marchi. Recommender
systems in mobile apps for health a systematic review. In Iberian Conference on Information Systems and
Technologies, CISTI , 2017.
[451] Weiwei Yuan, Chenliang Li, Donghai Guan, Guangjie Han, and Asad Masood Khattak. Socialized healthcare
service recommendation using deep learning. Neural Computing and Applications , 30(7), 2018.
[452] Hafed Zarzour, Ziad Al-Sharif, Mahmoud Al-Ayyoub, and Yaser Jararweh. A new collaborative filtering recom-
mendation algorithm based on dimensionality reduction and clustering techniques. In 2018 9th International
Conference on Information and Communication Systems, ICICS 2018 , volume 2018-January, 2018.
[453] Xiaoyi Deng and Feifei Huangfu. Collaborative Variational Deep Learning for Healthcare Recommendation.
IEEE Access , 7, 2019.
[454] Gagangeet Singh Aujla, Anish Jindal, Rajat Chaudhary, Neeraj Kumar, Sahil Vashist, Neeraj Sharma, and
Mohammad S. Obaidat. DLRS: Deep Learning-Based Recommender System for Smart Healthcare Ecosystem.
InIEEE International Conference on Communications , volume 2019-May, 2019.
61
Page 62:
arXiv Template A P REPRINT
[455] Abhaya Kumar Sahoo, Chittaranjan Pradhan, Rabindra Kumar Barik, and Harishchandra Dubey. DeepReco:
Deep learning based health recommender system using collaborative filtering. Computation , 7(2), 2019.
[456] Celestine Iwendi, Suleman Khan, Joseph Henry Anajemba, Ali Kashif Bashir, and Fazal Noor. Realizing an
Efficient IoMT-Assisted Patient Diet Recommendation System Through Machine Learning Model. IEEE Access ,
8, 2020.
[457] Satvik Garg. Drug recommendation system based on sentiment analysis of drug reviews using machine learning.
InProceedings of the Confluence 2021: 11th International Conference on Cloud Computing, Data Science and
Engineering , 2021.
[458] Maryam Al-Ghamdi, Hanan Elazhary, and Aalaa Mojahed. Evaluation of Collaborative Filtering for Recom-
mender Systems. International Journal of Advanced Computer Science and Applications , 12(3), 2021.
[459] Cecilia S. Lee and Aaron Y . Lee. Clinical applications of continual learning machine learning, 2020.
[460] Anam Mustaqeem, Syed Muhammad Anwar, and Muhammad Majid. A modular cluster based collaborative
recommender system for cardiac patients. Artificial Intelligence in Medicine , 102, 2020.
[461] Shaina Raza and Chen Ding. Improving Clinical Decision Making with a Two-Stage Recommender System.
IEEE/ACM Transactions on Computational Biology and Bioinformatics , 2023.
[462] Luis F. Luna-Reyes and J. Ramon Gil-Garcia. Digital government transformation and internet portals: The
co-evolution of technology, organizations, and institutions. Government Information Quarterly , 31(4), 2014.
[463] Jie Lu, Qusai Shambour, Yisi Xu, Qing Lin, and Guangquan Zhang. BizSeeker: A hybrid semantic recommenda-
tion system for personalized government-to-business e-services. Internet Research , 20(3), 2010.
[464] Xuetao Guo and Jie Lu. Intelligent e-Government services with personalized recommendation techniques.
International Journal of Intelligent Systems , 22(5), 2007.
[465] Ninghua Sun, Tao Chen, Wenshan Guo, and Longya Ran. Enhanced collaborative filtering for personalized
e-government recommendation. Applied Sciences (Switzerland) , 11(24), 2021.
[466] Ninghua Sun, Qiangqiang Luo, Longya Ran, and Peng Jia. Similarity matrix enhanced collaborative filtering for
e-government recommendation. Data and Knowledge Engineering , 145, 2023.
[467] Bahrudin Hrnjica, Denis Music, and Selver Softic. Model-Based Recommender Systems. EAI/Springer
Innovations in Communication and Computing , pages 125–146, 2020.
[468] Ninghua Sun, Tao Chen, Qiangqiang Luo, and Longya Ran. User dynamic topology-information-based matrix
factorization for e-government recommendation. Applied Soft Computing , 124, 2022.
[469] Alhassan Jamilu Ibrahim, Peter Zira, and Nuraini Abdulganiyyi. Hybrid Recommender for Research Papers and
Articles. International Journal of Intelligent Information Systems , 10(2), 2021.
[470] Folasade Olubusola Isinkaye and Tomiwa John Fred-Yusuff. An E-Library System Integrated with Bookshelf
and Recommendation Components. Journal of Applied Intelligent System , 7(1), 2022.
[471] Jieun Son and Seoung Bum Kim. Academic paper recommender system using multilevel simultaneous citation
networks. Decision Support Systems , 105, 2018.
[472] Shutian Ma, Chengzhi Zhang, and Xiaozhong Liu. A review of citation recommendation: from textual content to
enriched context. Scientometrics , 122(3), 2020.
[473] Qian Zhang, Jie Lu, and Guangquan Zhang. Recommender Systems in E-learning. Journal of Smart Environments
and Green Computing , 1(2):76–89, 4 2021.
[474] Shini Renjith, A. Sreekumar, and M. Jathavedan. An extensive study on the evolution of context-aware
personalized travel recommender systems. Information Processing and Management , 57(1), 2020.
[475] Francesco Ricci. Recommender Systems in Tourism. Handbook of e-Tourism , pages 1–18, 2020.
[476] Booking.com: How we work.
[477] Si Shi, Yuhuang Gong, and Dogan Gursoy. Antecedents of Trust and Adoption Intention toward Artificially
Intelligent Recommendation Systems in Travel Planning: A Heuristic–Systematic Model. Journal of Travel
Research , 60(8), 2021.
[478] Miguel Torres-Ruiz, Felix Mata, Roberto Zagal, Giovanni Guzmán, Rolando Quintero, and Marco Moreno-Ibarra.
A recommender system to generate museum itineraries applying augmented reality and social-sensor mining
techniques. Virtual Reality , 24(1), 2020.
[479] Jun Chang, Wenting Tu, Changrui Yu, and Chuan Qin. Assessing dynamic qualities of investor sentiments for
stock recommendation. Information Processing and Management , 58(2), 2021.
62
Page 63:
arXiv Template A P REPRINT
[480] Lawrence Bunnell, Kweku Muata Osei-Bryson, and Victoria Y . Yoon. FinPathlight: Framework for an multiagent
recommender system designed to increase consumer financial capability. Decision Support Systems , 134, 2020.
[481] Mao Kang, Ye Bi, Zhenyu Wu, Jianming Wang, and Jing Xiao. A heterogeneous conversational recommender
system for financial products. In CEUR Workshop Proceedings , volume 2601, 2020.
[482] Shivani Bharatbhai Patel, Pronaya Bhattacharya, Sudeep Tanwar, and Neeraj Kumar. KiRTi: A Blockchain-Based
Credit Recommender System for Financial Institutions. IEEE Transactions on Network Science and Engineering ,
8(2), 2021.
[483] Asefeh Asemi, Adeleh Asemi, and Andrea Ko. Unveiling the impact of managerial traits on investor decision
prediction: ANFIS approach. Soft Computing , 2023.
[484] Qi Zhang, Jingjie Li, Qinglin Jia, Chuyuan Wang, Jieming Zhu, Zhaowei Wang, and Xiuqiang He. UNBERT:
User-News Matching BERT for News Recommendation. In IJCAI International Joint Conference on Artificial
Intelligence , 2021.
[485] Shitao Xiao, Zheng Liu, Yingxia Shao, Tao Di, Bhuvan Middha, Fangzhao Wu, and Xing Xie. Training Large-
Scale News Recommenders with Pretrained Language Models in the Loop. In Proceedings of the ACM SIGKDD
International Conference on Knowledge Discovery and Data Mining , 2022.
[486] Mykola Makhortykh and Mariëlle Wijermars. Can Filter Bubbles Protect Information Freedom? Discussions of
Algorithmic News Recommenders in Eastern Europe. Digital Journalism , 11(9), 2023.
[487] Chuhan Wu, Fangzhao Wu, Xiting Wang, Yongfeng Huang, and Xing Xie. FairRec: Fairness-aware News
Recommendation with Decomposed Adversarial Learning. In 35th AAAI Conference on Artificial Intelligence,
AAAI 2021 , volume 5B, 2021.
[488] Paul Covington, Jay Adams, and Emre Sargin. Deep neural networks for youtube recommendations. In RecSys
2016 - Proceedings of the 10th ACM Conference on Recommender Systems , 2016.
[489] Krishnaram Kenthapadi, Benjamin Le, and Ganesh Venkataraman. Personalized job recommendation system at
LinkedIn: Practical challenges and lessons learned. In RecSys 2017 - Proceedings of the 11th ACM Conference
on Recommender Systems , 2017.
[490] Rahul Katarya and Yamini Arora. A Survey of Recommendation Systems in Twitter. In International Conference
on "Computational Intelligence and Communication Technology", CICT 2018 ,
2018.
[491] Weihao Gao, Xiangjun Fan, Chong Wang, Jiankai Sun, Kai Jia, Wenzhi Xiao Ruofan Ding, Xingyan Bin, Hui
Yang, Xiaobing Liu ByteDance Inc, Wenzhi Xiao, Ruofan Ding, and Xiaobing Liu. Deep Retrieval: Learning A
Retrievable Structure for Large-Scale Recommendations; Deep Retrieval: Learning A Retrievable Structure for
Large-Scale Recommendations. 2021.
[492] Zhuoran Liu, Leqi Zou, Xuan Zou, Caihua Wang, Biao Zhang, Da Tang, Bolin Zhu, Yijie Zhu, Peng Wu,
Ke Wang, and Youlong Cheng. Monolith: Real Time Recommendation System With Collisionless Embedding
Table. In CEUR Workshop Proceedings , volume 3303, 2022.
[493] Shiwen Zhao, Charles Crissman, and Guillermo R Sapiro. Consistent Collaborative Filtering via Tensor
Decomposition. 1 2022.
[494] Mark Levy Apple, Bruno Di, Giorgi Apple, Floris Weers Apple, Angelos Katharopoulos, and Tom Nickson Apple.
Controllable Music Production with Diffusion Models and Guidance Gradients. 11 2023.
[495] Shashank Rajput, Nikhil Mehta, Deepmind Google, Anima, Singh Google, Deepmind Raghunandan, Keshavan
Google, Trung Vu, Google Lukasz, Heldt Google Lichan, Hong Google Deepmind, Yi Tay, Google Deepmind,
Vinh Q Tran Google, Jonah Samost, Google Maciej, Kula Google Deepmind, Ed H Chi Google, Deepmind
Maheswaran, and Sathiamoorthy Google Deepmind. Recommender Systems with Generative Retrieval. 5 2023.
[496] S. Bhaskaran and Raja Marappan. Design and analysis of an efficient machine learning based hybrid recommen-
dation system with enhanced density-based spatial clustering for digital e-learning applications. Complex and
Intelligent Systems , 9(4), 2023.
[497] Mounia Rahhali, Lahcen Oughdir, Youssef Jedidi, Youssef Lahmadi, and Mohammed Zakariae El Khattabi.
E-learning Recommendation System Based on Cloud Computing. In Lecture Notes in Electrical Engineering ,
volume 745, 2022.
[498] Magdalini Eirinaki, Jerry Gao, Iraklis Varlamis, and Konstantinos Tserpes. Recommender systems for large-scale
social networks: A review of challenges and solutions, 2018.
[499] Monika Singh. Scalability and sparsity issues in recommender datasets: a survey. Knowledge and Information
Systems , 62(1):1–43, 2020.
63
Page 64:
arXiv Template A P REPRINT
[500] Zeshan Fayyaz, Mahsa Ebrahimian, Dina Nawara, Ahmed Ibrahim, and Rasha Kashef. Recommendation
systems: Algorithms, challenges, metrics, and business opportunities. applied sciences , 10(21):7748, 2020.
[501] Zefeng Chen, Wensheng Gan, Jiayang Wu, Kaixia Hu, and Hong Lin. Data scarcity in recommendation systems:
A survey. ACM Transactions on Recommender Systems , 2024.
[502] Ninghao Liu, Yong Ge, Li Li, Xia Hu, Rui Chen, and Soo Hyun Choi. Explainable Recommender Systems via
Resolving Learning Representations. In International Conference on Information and Knowledge Management,
Proceedings , 2020.
[503] Dongmin Hyun, Chanyoung Park, Min Chul Yang, Ilhyeon Song, Jung Tae Lee, and Hwanjo Yu. Review
sentiment-guided scalable deep recommender system. In 41st International ACM SIGIR Conference on Research
and Development in Information Retrieval, SIGIR 2018 , 2018.
[504] Dietmar Jannach and Michael Jugovac. Measuring the business value of recommender systems. ACM Transac-
tions on Management Information Systems (TMIS) , 10(4):1–23, 2019.
[505] Christoph Trattner, Dietmar Jannach, Enrico Motta, Irene Costera Meijer, Nicholas Diakopoulos, Mehdi Elahi,
Andreas L Opdahl, Bjørnar Tessem, Njål Borch, Morten Fjeld, et al. Responsible media technology and ai:
challenges and research directions. AI and Ethics , 2(4):585–594, 2022.
[506] Yashar Deldjoo and Tommaso Di Noia. Cfairllm: Consumer fairness evaluation in large-language model
recommender system. arXiv preprint arXiv:2403.05668 , 2024.
[507] Chen Gao, Yu Zheng, Wenjie Wang, Fuli Feng, Xiangnan He, and Yong Li. Causal inference in recommender
systems: A survey and future directions. ACM Transactions on Information Systems , 42(4):1–32, 2024.
[508] Hongzhi Yin, Liang Qu, Tong Chen, Wei Yuan, Ruiqi Zheng, Jing Long, Xin Xia, Yuhui Shi, and Chengqi Zhang.
On-device recommender systems: A comprehensive survey. arXiv preprint arXiv:2401.11441 , 2024.
[509] Xin Xia, Junliang Yu, Qinyong Wang, Chaoqun Yang, Nguyen Quoc Viet Hung, and Hongzhi Yin. Efficient
on-device session-based recommendation. ACM Transactions on Information Systems , 41(4):1–24, 2023.
[510] Shaina Raza, Shardul Ghuge, Chen Ding, Elham Dolatabadi, and Deval Pandya. Fair enough: Develop and
assess a fair-compliant dataset for large language model training? Data Intelligence , 6(2):559–585, 2024.
Appendix
Evaluation Criteria
In this paper, we labeled each reviewed paper based on the following criteria:
•Scalability : A paper was labeled as high, medium, or low scalability based on the system’s ability to handle
increasing amounts of data and users. High scalability indicates the system can efficiently manage large-scale
data and user bases, medium scalability indicates moderate efficiency, and low scalability indicates limited
capability in scaling up.
•Interpretability : This attribute was labeled high, medium, or low depending on how easily the system’s
recommendations can be understood by users. High interpretability means the system’s outputs are easily
explainable, medium interpretability means some effort is needed to understand the recommendations, and low
interpretability means the system’s logic is complex and not easily understandable.
•Computational Efficiency : We assessed this by measuring the system’s ability to provide recommendations
quickly and with minimal computational resources. High efficiency means the system operates swiftly with
low resource usage, medium efficiency indicates moderate performance, and low efficiency means the system
requires significant computational resources and time.
•Reproducibility : Papers were labeled based on how consistently the system’s results can be replicated under
the same conditions. High reproducibility means the experiments can be consistently reproduced, medium
reproducibility indicates some variations might occur, and low reproducibility means significant discrepancies
are likely when the experiments are repeated.
List of Tables
1 Overview of Related Surveys Ordered by Date of Publication and Comparison Criteria . . . . . . . . 3
2 Inclusion and Exclusion Criteria for Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . 6
64
Page 65:
arXiv Template A P REPRINT
3 Identified Papers by Database . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4 Publications on Foundational RS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
5 Deep Learning-based RS Publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
6 Comprehensive Overview of Graph Neural Network Models across Various Metrics and Use Cases. This
table details each model’s Input features, Year of Publication, and Characteristics such as Scalability,
Interpretability, Efficiency, and Reproducibility (rated as High, Medium, or Low, the symbol ’-’ means
no information available for this). It also lists the Dataset Used, Evaluation Metrics, Model Accuracy
(as per evaluation metric from the previous column), Learning Task, and Application Field. . . . . . . 14
7 Sequential Models. This table provides a detailed overview of various sequential models in recommen-
dation systems, showcasing their combined characteristics of Scalability, Interpretability, Computational
Efficiency, and Reproducibility (rated as High, Medium, or Low). Additionally, the table includes
information on datasets used, evaluation metrics, model accuracy, publication year, and application fields. 16
8 Comprehensive Overview of Knowledge Graph Based Recommender System Models across Various
Metrics and Use Cases. This table details each model’s Input features, Year of Publication, and
Characteristics such as Scalability, Interpretability, Efficiency, and Reproducibility (rated as High,
Medium, or Low). It also lists the Dataset Used, Evaluation Metrics, Model Accuracy, Learning Task,
and Application Field. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
9 Comprehensive Overview of Reinforcement Learning based Recommender System Models across
Various Metrics and Use Cases. This table details each model’s Input features, Year of Publication,
and Characteristics such as Scalability, Interpretability, Efficiency, and Reproducibility (rated as High,
Medium, or Low). It also lists the Dataset Used, Evaluation Metrics, Model Accuracy, Learning
Task, and Application Field. Metrics that their numerical value is not reported are specified with “No
numerical value”. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
10 Comprehensive Overview of LLM Based Models across Various Metrics and Use Cases. This table
details each model’s Input features, Year of Publication, and Characteristics such as Scalability, Inter-
pretability, Efficiency, and Reproducibility (rated as High, Medium, or Low). It also lists the Dataset
Used, Evaluation Metrics, Model Accuracy, Learning Task, and Application Field. . . . . . . . . . . 23
11 Comprehensive Overview of Multi-Modal Based Models across Various Metrics and Use Cases. This
table details each model’s Input features, Year of Publication, and Characteristics such as Scalability,
Interpretability, Efficiency, and Reproducibility (rated as High, Medium, or Low). It also lists the
Dataset Used, Evaluation Metrics, Model Accuracy, Learning Task, and Application Field. . . . . . . 26
12 Comprehensive Overview of Context-Aware Recommender System Models across Various Metrics and
Use Cases. This table details each model’s Input features, Year of Publication, and Characteristics such
as Scalability, Interpretability, Efficiency, and Reproducibility (rated as High, Medium, or Low). It also
lists the Dataset Used, Evaluation Metrics, Model Accuracy, Learning Task, and Application Field. . . 28
13 Comprehensive Overview of Review Based Models across Various Metrics and Use Cases. This
table details each model’s Input features, Year of Publication, and Characteristics such as Scalability,
Interpretability, Efficiency, and Reproducibility (rated as High, Medium, or Low). It also lists the
Dataset Used, Evaluation Metrics, Model Accuracy, Learning Task, and Application Field. . . . . . . 29
14 Comprehensive Overview of Aspect Based Models across Various Metrics and Use Cases. This
table details each model’s Input features, Year of Publication, and Characteristics such as Scalability,
Interpretability, Efficiency, and Reproducibility (rated as High, Medium, or Low). It also lists the
Dataset Used, Evaluation Metrics, Model Accuracy, Learning Task, and Application Field. Metrics that
their numerical value is not reported are specified with “No numerical value”. . . . . . . . . . . . . . 31
15 Comprehensive Overview of Explainable and Trustworthy Recommender System Models across
Various Metrics and Use Cases. This table details each model’s Input features, Year of Publication,
and Characteristics such as Scalability, Interpretability, Efficiency, and Reproducibility (rated as High,
Medium, or Low). It also lists the Dataset Used, Evaluation Metrics, Model Accuracy, Learning Task,
and Application Field. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
65
Page 66:
arXiv Template A P REPRINT
16 Comprehensive Overview Recommender System Models for FATE across Various Metrics and Use
Cases. This table details each model’s Input features, Year of Publication, and Characteristics such as
Scalability, Interpretability, Efficiency, and Reproducibility (rated as High, Medium, or Low). It also
lists the Dataset Used, Evaluation Metrics, Model Accuracy, Learning Task, and Application Field. . . 34
17 Publications by Industry in Recommendation Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 37
66
