Available online at www.sciencedirect.

com

ScienceDirect
Procedia CIRP 121 (2024) 115–120

11th CIRP Global Web Conference (CIRPe 2023)

Unveiling the Latest Trends and Advancements in Machine Learning

Algorithms for Recommender Systems: A Literature Review
Sara Shafieea,*
a
Technical University of Denmark, Department of Mechanical and Civil Engineering, 2800 Kgs. Lyngby, Denmark

* Corresponding author. Tel.: +45 50304495. E-mail address: [email protected]

Abstract

This paper presents a comprehensive literature review of the research and application of machine learning (ML) algorithms in recommender
systems (RS). The study aims to identify recent trends, explore real-life applications, and guide researchers in positioning their research activities
in this domain published in 2023 (Jan-June). The findings are categorized into different domains including education, healthcare, ML algorithms
(auto-encoders and reinforcement learning), e-commerce, and digital journalism. The review highlights the enhanced recommendation accuracy,
increased scalability, personalization and context awareness, diverse ML techniques, and strategies for handling cold start and data sparsity, and
the foundation for future advancements in ML algorithms for RSs considering the application in manufacturing enterprises.
© 2023 The Authors. Published by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0)
Peer-review under responsibility of the scientific committee of the 11th CIRP Global Web Conference
Keywords: Machine learning; Recommender system (RS); Personalization; Review; Manufacturing; Scalability

1. Introduction aggregated characteristics of individual user and item

information [4]. These features encompass various aspects,
Recommender systems (RS) have widely adopted in various including a user's demographic attributes, interests, item
online applications, enabling users to suggest relevant specifications, transaction contexts, and temporal usage
products, services, and content [1]. RSs are designed to suggest patterns that reflect user behavior and preferences [5]. As the
items or content to users based on their preferences and field of RSs evolved, researchers began exploring the use of
interests. These systems can be found in various domains such machine learning (ML) algorithms to enhance recommendation
as e-commerce, streaming platforms, social media, etc. Today, accuracy and address the limitations of traditional approaches
RSs are used in many information-based large companies such [6]. ML, a subfield of artificial intelligence (AI), focuses on
as Google, Twitter, LinkedIn, and Netflix [2]. There are a few developing algorithms that can automatically learn and
studies investigating the application of RSs in manufacturing. improve from data [7]. Here are some key points highlighting
For example, Yao et al. [3] proposed hybrid machine learning the integration of ML algorithms in RSs:
method to provide feasible conceptual design solutions for A) Enhanced Recommendation Accuracy: ML algorithms
inexperienced designers by recommending appropriate bring the potential to capture complex patterns and
additive manufacturing design features through a case study. relationships in user-item interactions, leading to more
Previous studies in recommendation algorithms commonly accurate recommendations [8]. ML algorithms can learn
employ two types of features: collective local features and from historical data, user feedback, and contextual
graph-related features. Collective local features capture the information to make personalized recommendations.

2212-8271 © 2023 The Authors. Published by Elsevier B.V.

This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0)
Peer-review under responsibility of the scientific committee of the 11th CIRP Global Web Conference
10.1016/j.procir.2023.08.062
116 Sara Shafiee / Procedia CIRP 121 (2024) 115–120

B) Increased Scalability and Flexibility: ML algorithms can as configuration systems [15], rely on various techniques, such
efficiently handle large-scale datasets, making them suitable as rule-based reasoning, ontologies, knowledge graphs,
for RS with millions of users and items [9]. They offer semantic analysis, and natural language processing, to extract
flexibility in modeling various aspects of user preferences, and represent knowledge about users and items [16].
item characteristics, and contextual factors. Demography-based RSs are a specialized class of
C) Personalization and Context Awareness: ML algorithms recommendation systems that leverage demographic
enable personalized recommendations by learning user information to generate personalized user recommendations
preferences and tailoring suggestions to individual users [17,18]. A community-based RS is a type of RSs that utilizes
[10]. Context-aware RSs consider situational factors such as the collective wisdom and opinions of a community of users to
time, location, and device to provide more relevant and generate personalized recommendations [19,20]. ML
timely recommendations [11]. algorithms can be classified into four main categories:
D) Diverse ML Techniques: ML algorithms play a key role in supervised, unsupervised, semi-supervised, and reinforcement
enhancing RSs with different applications. Decision trees, learning [2].
random forests, and gradient-boosting methods are used for
content-based filtering, while neural networks and CNNs
capture complex item features and user representations.
Reinforcement learning algorithms further contribute to
optimal recommendation strategies. Integration of these
diverse ML techniques improves RSs by enhancing
recommendation accuracy and personalization [2].
E) Handling Cold Start and Data Sparsity: ML algorithms offer
strategies to mitigate the cold-start problem, where limited
data is available for new users or items. Techniques like
matrix completion, transfer learning, and feature
engineering can be employed to handle data sparsity and
make recommendations in such scenarios [2,12].
F) Revolutionize Manufacturing Enterprises: ML algorithms
can optimize production processes, supply chain
management, maintenance planning, and product quality Fig. 1. The main categories of recommendation systems
control [3].
The objective of this paper is to provide researchers and Table 1. Literature review papers distribution in machine learning algorithms
for recommender systems (2023)
practitioners with valuable insights into the RS field and the
utilization of ML techniques. The study focuses on three main Main topics Available literature Publication channel
papers
goals: (i) identifying recent trends in the use and research of
ML algorithms in RSs, (ii) exploring real-life applications of Educational RS (4) [21–24] Applied Artificial Intelligence; IEEE
Access; International Journal of
RSs employing ML algorithms, and (iii) guiding new
Engineering Trends and Technology;
researchers positioning their research activities in this domain. Education and Information Technologies

Health RS (3) [25–27] Journal of Medical Internet Research;

2. Background and methods International Journal of Environmental
Research and Public Health; Expert
RSs can be categorized into four main approaches (see Fig. Systems with Applications

1): collaborative filtering, content-based filtering, hybrid ML algorithms in RS (2) [28,29] Mathematics; Energies
filtering, and other personalized services. Collaborative E-Commerce RS (1) [30] Applied Sciences (Switzerland)
filtering: These systems analyze user data to identify
Digital Journalism RS (1) [31] Expert Systems with Applications
similarities among users and recommend items based on the
preferences of similar users [13]. For example, in an online
music store, the system suggests songs purchased by users with Given the abundance of studies and considerable research
similar music tastes. Content-based filtering: RSs using this interest in the field of ML application in RSs and considering
approach focus on item data and recommend items with similar the limitations in the scope and space of this paper, our focus
attributes to the ones the user is interested in [11]. For instance, will be solely on the latest journal review papers published in
when a user searches for a computer online, the system 2023 (Jan-June). This review research has been conducted in
suggests computers with similar specifications and features. May 2023. To ensure a systematic approach, this paper adopts
Hybrid filtering: These systems combine both collaborative a three-step protocol consisting of data search, data analysis,
and content-based filtering techniques [14]. They consider both and report generation, forming the basis of our comprehensive
user and item data to generate recommendations. For example, literature review [32]. To guide the examination, the review
on a social network, the system may recommend profiles addresses the following three research questions (RQs): (1)
similar to the user's interests (collaborative filtering) and then What are the current trends in the use and research of RS when
use the recommended profiles to find new similar profiles implementing ML algorithms? (2) What are the emerging
(content-based filtering). Knowledge-based RSs, also known trends in the use and research of ML algorithms when
developing RS?
 Sara Shafiee / Procedia CIRP 121 (2024) 115–120 117

By addressing these research questions, the review aims to need for further research on the power and performance of
provide insights into the trends, developments, and sources of different online attributes. Different recommendation
research related to the utilization of ML algorithms in RS. The algorithms, including rule-based and collaborative filtering,
data search encompassed through the search string TITLE- were commonly used, but there is potential for exploring
ABS-KEY (("recommendation system" OR "recommender hybrid and ensemble algorithms for greater accuracy. These
system") AND "machine learning") was applied to search for findings contribute to the enhancement of e-learning systems
journal literature review articles in English on the Elsevier and classification models such as decision trees, logistic
Scopus database. Due to a significant number of literature, we regression, or support vector machines, are often employed to
limited the search only to 2023 which resulted in 189 journal classify learner attributes, preferences, or learning styles,
papers. Next, the abstracts of the identified publications were enabling personalized and tailored recommendations.
scanned to determine the initial sample of the relevant Pal et al. [23] screened 40 papers to identify RSs that
literature. Based on the selection criteria, which explicitly specifically support the teaching/learning environment. The
addressed the ML application for RS as the main topic of study analysis focused on different techniques and approaches used
and contributions that explicitly addressed the scope of the for implementation, as well as evaluation measures employed
research (e.g., reviews, definitions, frameworks), a subset of 11 to assess the quality and accuracy of the recommendation
articles was selected for full-text reading. framework. The following techniques are discussed: (1) Deep
Neural Networks: Using deep neural networks to recommend
3. Literature review results resources based on learners' interests; (2) User Feedback:
Incorporating user feedback; (3) Content-Based and
3.1. Research perspectives and opportunities in Educational Collaborative Filtering: Combining content-based and
Recommender Systems collaborative filtering in the LeCoRe approach; (4) Web-Based
Personalized Recommendations: Leveraging past ratings and
The amount of data generated by educational institutions is
attention-based CNN techniques for personalized
used to make recommendations for a wide range of areas [21].
recommendations on MOOCs platforms; (5) Multi-
Maphosa and Maphosa [21] reviewed 272 articles published
Personalized RS for E-Learning: Considering learners'
between 2007 and 2021. Research indicates that many college
preferences, interests, and history through content-based and
students enter their studies without a clear career path and often
collaborative filtering; (6) Learners' Annotation Activities:
change their majors’ multiple times. RSs have assisted students
Utilizing learners' activity annotations for personalized
in selecting their majors using techniques like the nearest
recommendations; (7) Hybrid Filtering: Integrating content-
neighbor algorithm. Educational RS utilizes collaborative
based, collaborative filtering, and hybrid filtering approaches;
filtering and nearest-neighbor algorithms to rank and
(8) LSTM and RNN: Utilizing LSTM and RNN deep learning
recommend items based on user ratings and preferences.
techniques for sequence-based course recommendations.
Content-based recommendations and data mining techniques
Future research in this field can focus on refining
are also used to generate personalized learner
recommendation algorithms, developing standardized
recommendations. RS in education uses collaborative filtering
evaluation measures, utilizing diverse educational datasets,
and content-based techniques to recommend learning materials
integrating intelligent tutoring systems and adaptive learning
and courses to students. RS also assists in selecting suitable
technologies, and enhancing the accuracy and personalization
majors and provides teaching patterns for educators. Future
of RSs in the context. These advancements will contribute to
research can explore integrating bibliometric analysis and the
more effective and engaging educational experiences for
PRISMA methodology to provide a comprehensive overview
learners.
of RSs research in higher education. This approach can provide
da Silva et al. [24] studies 15 literature review studies in the
valuable insights into the current status and future trends of RS
field of evaluating educational RS (ERS). Evaluation
in this domain, benefiting researchers, policy-makers, and
methodologies for ERSs include offline experiments, user
practitioners. Furthermore, there is a need for more research on
studies, and online experiments. User studies and offline
RS in higher education from developing countries, as AI
experiments are commonly used, while real-life tests are less
applications such as RS can significantly impact student
frequent due to cost and complexity. User satisfaction
learning outcomes.
assessment is common in user studies, often combined with
Thongchotchat et al. [22] analyzed 40 papers on RSs
other experiments. Accuracy investigation (to measure the
utilizing learning styles. The systematic literature studies
effectiveness and alignment between the recommendation and
revealed advancements, prospects, and obstacles in educational
the customer's desired option) is the most prevalent research
RS, particularly in utilizing learning styles for
goal. There is a need for more real-life tests to understand
recommendations. Most systems focused on attribute-related
recommender impact. Offline experiments use larger datasets,
recommendations, while research in educational
while user studies involve smaller samples. Future research
recommendations and learning style theories requires further
should focus on real-life tests, evaluating pedagogical
investigation and growth. The analysis also highlighted the
effectiveness, and exploring diverse evaluation approaches.
need for assessing and comparing RSs based on different
Research opportunities in the field of ERSs have been
learning theories to enhance learning outcomes. Combining
identified based on the gaps observed in the analyzed papers.
theories of learning styles can enhance flexibility and
Three main strands of research opportunities were identified:
recommendation capacity. Questionnaires were the most
(1) Study of overlooked user attributes: existing ERS models
popular method for identifying learning styles, but there is a
118 Sara Shafiee / Procedia CIRP 121 (2024) 115–120

often neglect important user attributes such as emotional state handling uncertainties; and (6) Develop explainable
and cultural context; (2) Increase studies on ERS application in recommendation approaches to align with ethical guidelines.
informal learning situations: ERS research predominantly Etemadi et al. [27] identified, taxonomically classified, and
focuses on formal learning contexts, while the design and compared current HRS research systematically. Five sub-
evaluation of recommenders for informal learning settings are classes of applied approaches are identified, with collaborative-
overlooked; (3) Studies on the development of based approaches being the most common. The evaluation
multidimensional evaluation frameworks: existing evaluation factors in HRS primarily focus on accuracy, precision, recall,
of ERS primarily emphasizes accuracy and user satisfaction, and time, but scalability, security, and reliability should be
overlooking other critical factors such as student performance, addressed. Most articles use a real test-bed environment for
learning gains, and technological quality. The advancements in assessment, with Python being the most widely used
Educational RS have significant implications for the assessment tool at 15%. The reviewed papers ethical challenges
manufacturing industry, particularly in the domain of on-site related to privacy and data sharing, the proliferation of
training and learning factories. healthcare standards and the need for interoperability, the cold-
start problem in making recommendations for new items and
3.2. Research perspectives and opportunities in Health users, scalability challenges in handling a large amount of data,
Recommender Systems the importance of reliability in HRSs, the need for accuracy in
healthcare system results, patients and their relatives
According to Sun et al. [25], health RSs (HRS) are disagreement over treatment recommendations, authorization
specialized information retrieval systems that provide users and access to electronic medical records, convincing people to
with relevant health-related items based on their specific needs. trust and use healthcare systems, and legal aspects concerning
These systems aim to assist and motivate users in making the use of artificial intelligence in medicine. These challenges
informed decisions about their health and encourage behavior and research directions are crucial for advancing HRSs and
change. A total of 51 studies of HRSs were included in the require further exploration and consideration. Worker well-
systematic review and the health domains included general being is of paramount importance in the manufacturing
health promotion, lifestyle, generic health service, and some industry, especially in the context of Industry 5.0. Health RSs
other domains. Sun et al. [25] found that hybrid recommender offer valuable insights and recommendations for maintaining
systems, collaborative filtering, and knowledge-based systems worker health, which is crucial for ensuring a safe and
are commonly used in building HRS. Collaborative filtering productive manufacturing environment.
showed limited usage in health, with only a few studies
demonstrating its relevance. Some studies compared different 3.3. Research perspectives and opportunities in Auto-
recommender techniques to identify optimal algorithms. For Encoders and Reinforcement Learning
future research, Sun et al. [25] highlighted that it is important
to report and disclose detailed information. Further studies are Chen et al. [28] provided an overview of auto-encoder
needed to evaluate multiple domains, including clinical concepts, variants, applications, and their relationships with
effectiveness, patient perspectives, economic considerations, shallow and deep learning models. It explores the use of auto-
organizational aspects, sociocultural factors, and ethical and encoders in different fields with available toolkits. Auto-
legal considerations. It is also noted that there is a need for encoders gained popularity in unsupervised learning for their
more papers that combine clinical trials and simulations in the ability to process large amounts of unlabeled data. The survey
experimentation process, despite the potential benefits of such offers the examination of auto-encoder, covering theory,
an approach. features, perspectives, relationships with other models, toolkits,
Yera [26] reviewed 967 papers focused on food applications, and future trends. It explores potential areas such
recommendations for diabetic patients using the PRISMA 2020 as constructing hybrid models, incorporating attention
framework. Various research approaches were identified, mechanisms, integrating supervised learning, drawing
including ontologies, optimization methods, rule-based inspiration from neuroscience and cognitive science, and
systems, and user interaction. Challenges were found in optimizing parameter adjustment with improved algorithms
generalizing and incorporating diabetes-related knowledge, as (such as gradient descent or ADAM). Future studies in auto-
well as reproducing and generalizing results. Different encoders are suggested as (1) Hybrid Models: combining auto-
approaches such as optimization methods, data mining encoders with other models for improved performance; (2)
techniques, and interactive systems were identified, but Attention Mechanism: integrating visual attention mechanisms
limitations exist in terms of specific datasets, evaluation to enhance feature learning; (3) Supervised Learning
protocols, and generalization. The lack of a common Integration: using supervised learning to boost auto-encoder
framework and integration with AI-based methods pose performance. (4) Neuroscientific and Cognitive Science
challenges for future progress. To advance nutritional RSs for Approaches: aligning auto-encoder structures with biology and
diabetic patients, the following research directions have been cognition; (5) Optimization Algorithms: developing advanced
identified: (1) Establish a consolidated framework for further algorithms to automatically adjust model parameters.
research; (2) Better formalize the research problem and unify Sivamayil et al. [29] analyzed 127 publications focusing on
the definition and output of recommendations; (3) Utilize the applications of Reinforcement Learning (RL) in marketing,
domain knowledge from the medical and nutritional fields; (4) robotics, gaming, automated cars, natural language processing
Incorporate user preferences more effectively for personalized (NLP), internet of things (IOT) security, RSs, finance, and
recommendations; (5) Explore the use of fuzzy tools for energy management. RL is suitable for dynamic contexts due
 Sara Shafiee / Procedia CIRP 121 (2024) 115–120 119

to its ability to learn independently through trial and error findings into different clusters. Cluster 1 focused on
interactions. RL has the potential to revolutionize energy exploratory studies and ML approaches, exploring
management by optimizing energy usage in real-time and personalization, engagement metrics, and ML algorithms for
improving the functioning of renewable sources. While most news popularity prediction. Cluster 2 focused on text mining
energy management system applications are tested in and sentiment analysis, investigating fake news detection and
simulations, there is a need to implement them in real-world building reader confidence. Clusters 3 and 6 examined news
settings. RL-based RSs show high prediction accuracy, and RL recommendation and automated journalism, addressing ethical
is expanding its presence in gaming applications. In the issues and user perception. Clusters 4 and 5 focused on event
financial sector, RL is predominantly used in trading extraction and opinion mining, analyzing event dynamics and
applications, but there is room for exploration in credit risk reader comments. Future research can focus on personalization,
analysis. RL holds promise for enhancing the energy efficiency content automation, fact-checking, engagement metrics, and
of HVAC systems, although further research is needed to paywall mechanisms to enhance user experience, content
handle real-world complexity. Future studies in RL can explore optimization, information credibility, and business models in
algorithmic improvements, real-world implementations, digital journalism. Applying digital journalism RSs in the
transfer learning, robustness and safety, explainability, ethical manufacturing domain can facilitate effective communication
implications, multi-agent RL, and cross-disciplinary within the organization, enabling timely dissemination of
applications. These areas of research aim to enhance RL critical information, updates on production processes, and
algorithms, apply them in practical settings, address challenges, insights for decision-making at various levels of the
ensure safety and interpretability, consider ethical implications, manufacturing enterprise.
study multi-agent interactions, and explore diverse application
domains. Within the manufacturing industry, the application of 4. Concluding remarks
auto-encoders and reinforcement learning holds promise for
optimizing production line configurations, improving In this study, we aimed to address two main research
equipment setup based on historical data and real-time questions (RQs) to gain insights into the use and research of
performance metrics, and addressing the complexity of ML in RSs. The findings of this study indicate that the
manufacturing processes. integration of ML algorithms in various domains has shown
promising results in improving efficiency, accuracy, and
3.4. Research perspectives and opportunities in electronic- decision-making processes. The study’s objective was to
commerce recommender systems identify recent trends, explore real-life applications, and guide
researchers in this domain. The findings from this paper
Necula and Păvăloaia [30] investigated how AI is utilized in demonstrate that most review papers in this field (RQ1), in
RSs for electronic commerce. The review identified various 2023, are focused on the sectors of education [21–24] and then
AI-based approaches such as content-based scoring, healthcare [25–27]. We found two papers reviewing ML
collaborative filtering, deep learning, and virtual assistants. algorithms including Auto-Encoders and Reinforcement
These techniques aim to enhance recommendation relevance Learning [28,29]. We also found papers considering RS
and accuracy, improve the customer shopping experience, and applications in e-commerce [30] and journalism [31]. Turning
increase sales. The research direction in this field is moving to our RQ2 regarding the emerging trends in the use and
towards developing more sophisticated and adaptable AI-based research of ML algorithms in RSs, our review shed light on
RSs. The most common AI techniques used include ML, deep novel applications and advancements such as increasing
learning, and other relevant algorithms, which, when combined interest in the integration of reinforcement learning algorithms
with technologies like augmented reality, contribute to for optimal recommendation. The review has shed light on
improved relevance and personalization. Benefits include several aspects:
improved decision-making, reduced shopping effort, increased 1. Evolution of RSs: Over the years, RSs have evolved from
sales, and overcoming data issues. Further research is needed simple popularity-based approaches to more sophisticated
to focus on contexts where AI systems outperform traditional systems driven by ML algorithms. This evolution has been
ones. Research areas include sentiment analysis, embeddings, fueled by advancements in data collection, processing
and framework development. AI-based RSs can enhance e- power, and algorithmic techniques, enabling RSs to provide
commerce in further research. Integrating with other personalized and context-aware recommendations.
technologies like blockchain, IoT, and 5G networks should be 2. ML Algorithms' Contribution: ML algorithms have
explored. Customization and interaction in RSs, interplay significantly contributed to the success of modern RSs by
between various aspects like behavioral research are growing enabling the analysis of vast amounts of user and item data.
trends. The integration of e-commerce RSs with other 3. Challenges and Opportunities: Although ML algorithms
technologies like blockchain, IoT, and 5G networks opens up demonstrate significant potential, there are persistent
new avenues for enhancing manufacturing processes. challenges that need to be addressed, including data sparsity,
cold-start problems, and algorithmic bias. These challenges
3.5. Research Perspectives and Opportunities in Digital present opportunities for further research and innovation in
Journalism recommender systems the field of RSs, aiming to develop effective solutions.
4. Real-World Applications in Manufacturing Enterprises: The
Fernandes et al. [31] The study analyzed the role of Data
literature review has highlighted various real-world
Science (DS) in Digital Journalism (DJ) and categorized the
120 Sara Shafiee / Procedia CIRP 121 (2024) 115–120

applications where ML-based RSs have been successfully Frog Leaping Algorithm. International Journal of Intelligent Engineering
employed. By leveraging ML algorithms, RS can provide and Systems 2023;16:301–13. https://doi.org/10.22266/ijies2023.0228.27.
[15] Shafiee S, Kristjansdottir K, Hvam L, Forza C. How to scope
personalized recommendations for optimizing production configuration projects and manage the knowledge they require. J Knowl
processes, supply chain management, maintenance Manag 2018;22:982–1014. https://doi.org/10.1108/JKM-01-2017-0017.
planning, and product quality control. [16] Shafiee S, Wautelet Y, Friis SC, Lis L, Harlou U, Hvam L. Evaluating the
In manufacturing, RSs will optimize the production line by benefits of a computer-aided software engineering tool to develop and
recommending the most suitable equipment setup based on document product configuration systems. Comput Ind 2021;128:103432.
https://doi.org/10.1016/j.compind.2021.103432.
historical data, real-time performance metrics, and contextual [17] Yin S, Wang Y, Shafiee S. Ranking products through online reviews
information. There is a need for studies to focus on the considering the mass assignment of features based on BERT and -rung
application of RSs to analyze production data and predict the orthopair fuzzy set theory. Expert Syst Appl 2022:119142.
optimal configuration for achieving high productivity and https://doi.org/10.1016/J.ESWA.2022.119142.
minimizing downtime [33]. It is important to note that [18] Wang Y, Wu J, Lin L, Shafiee S. An online community-based dynamic
customisation model: the trade-off between customer satisfaction and
implementing ML-based RS in manufacturing enterprises enterprise profit. Int J Prod Res 2021;59:1–29.
requires careful consideration of data security, privacy, and https://doi.org/10.1080/00207543.2019.1693649.
transparency. Additionally, developing frameworks and [19] Wu J, Wang Y, Shafiee S, Zhang D. Discovery of associated consumer
guidelines for using ML-based RS in manufacturing enterprises demands: Construction of a co-demanded product network with
will ensure these systems' long-term sustainability. community detection. Expert Syst Appl 2021;178:115038.
https://doi.org/10.1016/j.eswa.2021.115038.
[20] Wang Y, Wu J, Zhang R, Shafiee S, Li C. A “user-knowledge-product”
References
Co-creation cyberspace model for product innovation. Complexity
2020;2020. https://doi.org/10.1155/2020/7190169.
[1] Huang Z, Hong C, Weng C, Lu WJ, Qu H. More Efficient Secure Matrix
[21] Maphosa V, Maphosa M. Fifteen Years of Recommender Systems
Multiplication for Unbalanced Recommender Systems. IEEE Trans
Research in Higher Education: Current Trends and Future Direction.
Dependable Secure Comput 2023;20:551–62.
Applied Artificial Intelligence 2023;37.
https://doi.org/10.1109/TDSC.2021.3139318.
https://doi.org/10.1080/08839514.2023.2175106.
[2] Portugal I, Alencar P, Cowan D. The use of machine learning algorithms
[22] Thongchotchat V, Kudo Y, Okada Y, Sato K. Educational
in recommender systems: A systematic review. Expert Syst Appl
Recommendation System Utilizing Learning Styles: A Systematic
2018;97:205–27. https://doi.org/10.1016/j.eswa.2017.12.020.
Literature Review. IEEE Access 2023;11:8988–99.
[3] Yao X, Moon SK, Bi G. A hybrid machine learning approach for additive
https://doi.org/10.1109/ACCESS.2023.3238417.
manufacturing design feature recommendation. Rapid Prototyp J
[23] Pal N, Dahiya O. A Systematic Literature Review on the Implications of
2017;23:983–97. https://doi.org/10.1108/RPJ-03-2016-0041.
Educational Recommender System in Teaching Learning Environment.
[4] Godinot A, Tarissan F. Measuring the effect of collaborative filtering on
International Journal of Engineering Trends and Technology 2023;71:203–
the diversity of users’ attention. Appl Netw Sci 2023;8.
12. https://doi.org/10.14445/22315381/IJETT-V71I4P218.
https://doi.org/10.1007/s41109-022-00530-7.
[24] da Silva FL, Slodkowski BK, da Silva KKA, Cazella SC. A systematic
[5] Kim HN, Ha I, Lee KS, Jo GS, El-Saddik A. Collaborative user modeling
literature review on educational recommender systems for teaching and
for enhanced content filtering in recommender systems. Decis Support Syst
learning: research trends, limitations and opportunities. Educ Inf Technol
2011;51:772–81. https://doi.org/10.1016/j.dss.2011.01.012.
(Dordr) 2023;28:3289–328. https://doi.org/10.1007/s10639-022-11341-9.
[6] Cai L, Lai T, Wang L, Zhou Y, Xiong Y. Graph convolutional network
[25] Sun Y, Zhou J, Ji M, Pei L, Wang Z. Development and Evaluation of
combining node similarity association and layer attention for personalized
Health Recommender Systems: Systematic Scoping Review and Evidence
recommendation. Eng Appl Artif Intell 2023;121.
Mapping. J Med Internet Res 2023;25. https://doi.org/10.2196/38184.
https://doi.org/10.1016/j.engappai.2023.105981.
[26] Yera R, Alzahrani AA, Martínez L, Rodríguez RM. A Systematic Review
[7] Li X, Chen H. Recommendation as link prediction in bipartite graphs: A
on Food Recommender Systems for Diabetic Patients. Int J Environ Res
graph kernel-based machine learning approach. Decis Support Syst
Public Health 2023;20. https://doi.org/10.3390/ijerph20054248.
2013;54:880–90. https://doi.org/10.1016/j.dss.2012.09.019.
[27] Etemadi M, Bazzaz Abkenar S, Ahmadzadeh A, Haghi Kashani M,
[8] Almas B, Mujtaba H, Khan KU. EHHR: an efficient evolutionary hyper-
Asghari P, Akbari M, et al. A systematic review of healthcare recommender
heuristic based recommender framework for short-text classifier selection.
systems: Open issues, challenges, and techniques. Expert Syst Appl
Cluster Comput 2023;26:1425–46. https://doi.org/10.1007/s10586-022-
2023;213. https://doi.org/10.1016/j.eswa.2022.118823.
03754-5.
[28] Chen S, Guo W. Auto-Encoders in Deep Learning—A Review with New
[9] Rani S, Kaur M, Kumar M, Ravi V, Ghosh U, Mohanty JR. Detection of
Perspectives. Mathematics 2023;11.
shilling attack in recommender system for YouTube video statistics using
https://doi.org/10.3390/math11081777.
machine learning techniques. Soft Comput 2023;27:377–89.
[29] Sivamayil K, Rajasekar E, Aljafari B, Nikolovski S, Vairavasundaram S,
https://doi.org/10.1007/s00500-021-05586-8.
Vairavasundaram I. A Systematic Study on Reinforcement Learning Based
[10] Zheng Y, Wei P, Chen Z, Cao Y, Lin L. Graph-Convolved Factorization
Applications. Energies (Basel) 2023;16.
Machines for Personalized Recommendation. IEEE Trans Knowl Data Eng
https://doi.org/10.3390/en16031512.
2023;35:1567–80. https://doi.org/10.1109/TKDE.2021.3100564.
[30] Necula SC, Păvăloaia VD. AI-Driven Recommendations: A Systematic
[11] Chaturvedi A, Aylward B, Shah S, Graziani G, Zhang J, Manuel B, et al.
Review of the State of the Art in E-Commerce. Applied Sciences
Content Recommendation Systems in Web-Based Mental Health Care:
(Switzerland) 2023;13. https://doi.org/10.3390/app13095531.
Real-world Application and Formative Evaluation. JMIR Form Res
[31] Fernandes E, Moro S, Cortez P. Data Science, Machine learning and big
2023;7. https://doi.org/10.2196/38831.
data in Digital Journalism: A survey of state-of-the-art, challenges and
[12] Kang L, Xue F, Hong R. RGCF: Refined graph convolution collaborative
opportunities. Expert Syst Appl 2023;221:119795.
filtering with concise and expressive embedding. Intelligent Data Analysis
https://doi.org/10.1016/j.eswa.2023.119795.
2022;26:427–45. https://doi.org/10.1145/nnnnnnn.nnnnnnn.
[32] de Almeida Biolchini JC, Mian PG, Natali ACC, Conte TU, Travassos
[13] Alshammari M, Alshammari A. Friend Recommendation Engine for
GH. Scientific research ontology to support systematic review in software
Facebook Users via Collaborative Filtering. International Journal of
engineering. Advanced Engineering Informatics 2007;21:133–51.
Computers, Communications and Control 2023;18.
https://doi.org/10.1016/J.AEI.2006.11.006.
https://doi.org/10.15837/ijccc.2023.2.4998.
[33] Wang C, Tan XP, Tor SB, Lim CS. Machine learning in additive
[14] Deenadayalan D, Kangaiammal A. User Feature Similarity Supported
manufacturing: State-of-the-art and perspectives. Addit Manuf 2020;36.
Collaborative Filtering for Page Recommendation Using Hybrid Shuffled
https://doi.org/10.1016/j.addma.2020.101538.