Intelligent Systems with Applications 24 (2024) 200435

Contents lists available at ScienceDirect

Intelligent Systems with Applications

journal homepage: www.journals.elsevier.com/intelligent-systems-with-applications

Artificial intelligence and recommender systems in e-commerce. Trends

and research agenda
Alejandro Valencia-Arias a,* , Hernán Uribe-Bedoya b, Juan David González-Ruiz c ,
Gustavo Sánchez Santos d, Edgard Chapoñan Ramírez e, Ezequiel Martínez Rojas f
a
School of Industrial Engineering, Universidad Señor de Sipán, Chiclayo 14001, Perú
b
Departamento de ciencias ambientales y de la construcción, Instituto Tecnológico Metropolitano, Medellín 050034, Colombia
c
Departamento de Economía, Grupo de Investigación en Finanzas y Sostenibilidad, Universidad Nacional de Colombia, Sede Medellín, Medellín 050034, Colombia
d
Universidad Ricardo Palma, Lima 15039, Perú
e
Escuela de Contabilidad, Universidad Señor de Sipán, Chiclayo 14001, Peru
f
Vicerrectoría de Investigación e Innovación, Universidad Arturo Prat, Chile, 1110939

A R T I C L E I N F O A B S T R A C T

Keywords: Combining recommendation systems and AI in e-commerce can improve the user experience and decision-
Advanced algorithms making. This study uses a method called bibliometrics to look at how these systems and artificial intelligence
User experience are changing. Of the 120 documents, 91 were analyzed. This shows a growth of 97.16% in the topic. The most
Sentiment analysis
influential authors were Paraschakis and Nilsson, with three publications and 43 citations. The magazine Elec­
Convolutional Neural Networks
PRISM-2020
tronic Commerce Research has four publications and 60 citations. China is the top country for citations, with
120, followed by India with 25 publications. The results show that research increased in 2021 and 2022. This
shows a shift towards sentiment analysis and convolutional neural networks. The identification of new keywords,
such as content-based image retrieval and knowledge graph, shows promising areas for future research. This
study provides a solid foundation for future research in e-commerce recommender systems.

1. Introduction the many products on e-commerce sites can confuse customers, making
it hard for them to find the best item.
The combination of recommender systems and AI in e-commerce has Recent research has explored AI techniques to address specific
led to a lot of research. There are different types of recommendations, challenges in recommender systems. For example, a recommender sys­
including virtual assistants and online suggestions made in real time. tem based on deep neural networks has been proposed to address the
These systems use AI to make e-commerce better for users and help them recommendation challenge in cold start cases on e-commerce platforms.
make decisions (Reyes, 2023). They analyze information to provide A deep neural network model was developed to suggest fashion com­
personalized recommendations. binations using additional e-commerce information (S. Wang & Qiu,
The growing availability of data and the complexity of user behavior 2021). A hierarchical recommendation system was introduced that in­
patterns has led to an increased demand for personalized recommen­ tegrates online user opinions to improve e-commerce recommendations
dations. To improve this, AI techniques have been added to recom­ (Islek & Oguducu, 2022).
mender systems to make predictions more accurate and deal with Also, researchers have studied ways to combine different AI tech­
problems like not having enough data and starting from scratch. Also, niques to improve recommendation systems for e-commerce. One

Abbreviations: CNNs, Convolutional Neural Networks; CNN, Convolutional Neural Network; GRU, Gated Recurrent Units; RNN, Recurrent Neural Network; RNNs,
Recurrent Neural Networks; WSDM, Web Search and Data Mining; SDAE, Sparse Denoising Autoencoder; LSTM, Long Short-Term Memory; IEEE, Institute of
Electrical and Electronics Engineers; AI, Artificial Intelligence; Variables, Adjustable Control Parameters; Number of publications, Search equations; Number of
citations, Inclusion/exclusion criteria; Temporal patterns of keyword usage, Selection thresholds; Keyword co-occurrence, Bibliometric indicators; Frequency of
keyword usage, Analysis tools (Microsoft Excel®, VOSviewer®); Validity of keyword usage, Data validation process.
* Corresponding author.
E-mail addresses: [email protected] (A. Valencia-Arias), [email protected] (H. Uribe-Bedoya), [email protected] (J.D. González-Ruiz),
[email protected] (E.C. Ramírez), [email protected] (E.M. Rojas).

https://doi.org/10.1016/j.iswa.2024.200435
Received 27 December 2023; Received in revised form 30 July 2024; Accepted 3 September 2024
Available online 6 September 2024
2667-3053/© 2024 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
A. Valencia-Arias et al. Intelligent Systems with Applications 24 (2024) 200435

example is a customer-focused system that uses a mix of sentiment 2020).

analysis (Karn et al., 2023). This means using AI to identify, extract, and This study will help e-commerce and AI by combining and expanding
analyze emotions, opinions, and attitudes expressed in online text existing knowledge. The study suggests using machine learning to un­
(Bawack, Wamba, Carillo & Akter, 2022; He & Liu, 2024). An unsu­ derstand how to use machine learning in e-commerce. The study shows
pervised machine learning method has been used to develop an image how to create effective recommendation systems. It explains how to use
recommendation system, which is similar to those used in e-commerce. advanced algorithms to improve the user experience and make recom­
This topic has generated a lot of interest in the research community. mendations more effective (Loukili, Messaoudi & El Ghazi, 2023).
These systems play an important role in improving user experience and Furthermore, the study provides a comprehensive overview of AI-driven
decision making in online business environments. Online recommender recommendations in e-commerce research (Necula & Păvăloaia, 2023).
systems are important and we study how they are used in e-commerce. The study aims to show how existing knowledge on recommendation
They help make personalised recommendations (Cabrera-Sánchez, systems for e-commerce can be improved. By understanding and
Ramos-de-Luna, Carvajal-Trujillo & Villarejo-Ramos, 2020). Despite addressing the problems found by other researchers, the study can help
recent recommendation systems being good at making recommenda­ improve and innovate. A study on online recommender systems by
tions, they have some problems, including being able to handle lots of Cabrera-Sánchez et al. (2020) looks at why they are used in e-commerce.
data, starting from scratch, and having lots of data (Roy & Dutta, 2022). Combining this with the proposed study could help identify new
The article also shows how artificial intelligence can help find natural research areas and focus points to improve recommender systems in
communities in complex networks, which is important for e-commerce online commercial environments. Similarly, an examination of AI in
(Chattopadhyay, Basu, Das, Ghosh & Murthy, 2021). These studies show recommendation systems provides an overview of the most recent trends
how recommender systems and artificial intelligence help e-commerce. and advancements in the field (Zhang, Lu & Jin, 2021). By combining
Anitha and Kalaiarasu (2021) show how collaborative machine this study with the proposed research, we can gain a more compre­
learning can improve e-commerce recommender systems. Their research hensive understanding of how AI techniques are transforming e-com­
shows how optimizing these methods helps provide more accurate and merce recommendation systems and how they can be effectively applied
personalized suggestions. The study also addresses how users interact to enhance user experience and increase recommendation efficiency.
with e-commerce platforms (H. Wang et al., 2019) and proposes a deep This article says that new ideas for e-commerce recommendations
neural network-based recommendation system to address the cold start and AI are needed to meet the growing demands and expectations. This
problem (H. Wang et al., 2019). approach is needed to overcome current research limitations and chal­
Artificial intelligence affects how features are chosen in recom­ lenges. Although there has been progress, there is still more to do. For
mender systems. The study highlights the importance of feature selec­ example, Acharya, Sassenberg and Soar (2023) say that we need to
tion based on conditional entropy in automated e-commerce understand how cognitive absorption affects people’s use of AI-based
recommender systems (Sharma & Sadagopan, 2022). This study shows recommender systems. Sahu and Gaur (2024) say that recommenda­
how AI can improve e-commerce recommender systems. tion systems in e-commerce should be more understandable. These
In e-commerce, intelligent search algorithms have improved effi­ studies show that we can understand and use AI in e-commerce better.
ciency and effectiveness. The gravitational search algorithm is one of This study is important because of this.
these systems that has proven useful in optimizing complex problems. A recent study by Shin et al. (2021) presents a new way to represent
This approach improves online recommendations by optimizing users in recommender systems for e-commerce. This study can help us
threshold settings. This improves user experience and recommendation understand how to make recommendations more personal and relevant
accuracy. It has also been used to identify dynamic systems in the to users. This will help us improve e-commerce strategies based on
context of electronic commerce. This method, inspired by classical me­ personalized recommendations.
chanics, helps improve recommendation systems. It makes recommen­ This research looks at how AI recommender systems are used in e-
dations more adaptable and personal to users (Mohammadi & Zahiri, commerce. It aims to set out a research agenda for future work. To this
2018). Integrating optimization into recommender systems makes them end, we propose the following research questions:
more adaptable and accurate.
Particle swarm optimization can solve many types of problems. It can When did interest in recommender systems and AI techniques in e-
help e-commerce systems work better. It can also help users get better commerce peak?
recommendations. Langs (2012) says this is a good approach. How many scientific articles on recommender systems and AI in e-
The effectiveness of using the inclined plane system algorithm to commerce have been published?
solve problems with multiple local optima was tested. The study shows What are the main research papers on recommender systems and AI
how inclined plane-based optimization can help with complex e-com­ in e-commerce?
merce challenges. They can also adapt recommendation strategies to suit What has changed in the topics studied in scientific articles on
different users in online shops (Darband, Mahboob, Shahbandegan & recommender systems and AI techniques in e-commerce?
Razavi, 2020). Integrating these advancements into intelligent optimi­ What are the main topics on recommender systems and AI techniques
zation algorithms can enhance the adaptability and accuracy of rec­ in e-commerce?
ommendations, which can improve customer satisfaction and loyalty in What are the new and growing keywords in recommender systems
e-commerce. and AI techniques in e-commerce?
Despite recent advances in recommender systems and AI in e-com­ What topics should be included in a research agenda on recom­
merce, there is still much to learn. A model has been created to predict mender systems and AI techniques in e-commerce?
how useful reviews are on e-commerce platforms. However, there is still
more to be done to improve the accuracy of recommendations The article starts with a look at how recommender systems and AI
(Almutairi & Abdullah, 2020). Although a hierarchical recommendation techniques are used in e-commerce. The Methodology section explains
system based on online user reviews has been introduced, it is essential how the study was done and how the data was collected and analyzed. It
to investigate how these techniques could be adapted and refined to deal answers the questions above. The results are in the Results section.
with the increasing complexity and diversity of data in e-commerce. Finally, the article explains what the results mean.
Also, while product ad recommendation in e-commerce has been
addressed using deep learning and distributed rendering techniques. 1.1. Motivations
However, it is important to investigate how these strategies could be
used more widely across different types of products and users (Zhou, This section looks at why researchers are studying recommender

2
A. Valencia-Arias et al. Intelligent Systems with Applications 24 (2024) 200435

systems and AI techniques in e-commerce. The reasons include the eliminating documents with incomplete indexing information. This en­
increasing use of e-commerce platforms, the need for personalised rec­ sures the reliability and coherence of the sources submitted to the bib­
ommendations to improve the user experience, and the potential of AI to liometric analysis.
improve recommendation strategies. It also discusses the gaps and
challenges in existing research that motivate further research in this 2.2. Source of information
field.
This study used Scopus and Web of Science as its main sources of
1.2. Innovations data. These databases are the best for scientific and academic research
because they cover a lot of relevant literature and have good analytical
This article looks at new ways of doing research on recommendation tools. Previous research has shown that Scopus and Web of Science are
systems and AI in e-commerce. It talks about new types of recommen­ reliable and representative sources for this bibliometric analysis (Zhu &
dation systems and algorithms that combine different AI techniques. It Liu, 2020).
also discusses new ways of making recommendations more accurate and
efficient. Finally, it looks at how AI can be used to solve specific prob­ 2.3. Search strategy
lems, such as problems with starting new sessions and user interaction.
Two search equations were developed to refine the search for in­
1.3. Contributions formation in the Scopus and Web of Science databases. These equations
were designed with the criteria for bibliometrics and the structure of the
This study assesses the impact of research in the field of electronic selected databases in mind. Specialized search equations are needed to
commerce and recommender systems. It consolidates existing knowl­ find studies on recommender systems and AI in e-commerce.
edge, expands research horizons, and addresses gaps in the existing For the Scopus database: (TITLE ("artificial intelligence" OR "deep
literature. Previous studies have contributed insights from bibliometric learning" OR "machine learning" OR "neural network algorithms" OR
analysis, recommendations for future research, and implications for "natural language processing" OR "computer vision techniques" OR
practitioners and stakeholders in the e-commerce industry. The pro­ "reinforcement learning") AND TITLE (e-commerce OR "electronic
posed study contributes to academic knowledge by explaining how AI- commerce" OR ecommerce OR "online commerce" OR "online store*" OR
powered recommender systems work in e-commerce. The article is "Internet commerce" OR "Internet store*" OR "virtual commerce" OR
structured as follows: "virtual shop*" OR "cyber commerce" OR "online commerce" OR "web
shop*") AND TITLE ("recommendation system*" OR "recommendation
1.4. Overview of article sections engines" OR "personalization system*")) OR (KEY ("artificial intelli­
gence" OR "deep learning" OR "machine learning" OR "neural network
The article includes the Abstract, which is a concise summary of the algorithms" OR "natural language processing" OR "computer vision" OR
research objec-tives, methods, findings, and conclusions. It also includes "reinforcement learning") AND TITLE (e-commerce OR "electronic
the Introduction, which sets the context for the study, outlines the commerce" OR ecommerce OR "online commerce" OR "online store*" OR
research problem, objectives, and significance, and provides an over­ "Internet commerce" OR "Internet store*" OR "virtual commerce" OR
view of the research approach. The methodology describes the research "virtual shop*" OR "cyber commerce" OR "online commerce" OR "web
design, data collection methods, analytical techniques, and any tools or shop*") AND KEY ("recommendation* system*" OR "recommendation*
soft-ware used for analysis. The results section presents the key findings engines" OR "personalization system*"))
of the study, in-cluding quantitative and qualitative data, visualizations, For the Web of Science database: (TI=("artificial intelligence" OR
and statistical analyses. The discussion section interprets the results in "deep learning" OR "machine learning" OR "neural network algorithms"
the context of existing literature, discussing implications, limitations, OR "natural language processing" OR "computer vision" OR "reinforce­
and future research directions, and providing insights into the broader ment learning") AND TI=(e-commerce OR "electronic commerce" OR
significance of the findings. Finally, the conclusions summarize the main ecommerce OR "online commerce" OR "online store*" OR "Internet
findings of the study, highlight contributions to the field, and offer commerce" OR "Internet store*" OR "virtual commerce" OR "virtual
concluding remarks on the research implications and potential impact. shop*" OR "cyber commerce" OR "online commerce" OR "web shop*")
AND TI=("recommendation system*" OR "recommendation engines" OR
2. Materials and methods "personalization system*")) OR (AK=("artificial intelligence" OR "deep
learning" OR "machine learning" OR "neural network algorithms" OR
This research follows the PRISMA-2020 guidelines, as presented in "natural language processing" OR "computer vision techniques" OR
Page et al. (2021). It looks at how recommender systems and AI tech­ "reinforcement learning") AND TI=(e-commerce OR "electronic com­
niques have changed over time and what they are used for in e-com­ merce" OR ecommerce OR "online commerce" OR "online store*" OR
merce. This approach will help us understand what has been done in this "Internet commerce" OR "Internet store*" OR "virtual commerce" OR
field and what still needs to be done. By using a good method, we can "virtual shop*" OR "cyber commerce" OR "online commerce" OR "web
give a good picture of what we know about this changing field. shop*") AND AK=("recommendation* system*" OR "recommendation*
engines" OR "personalization system*"))
2.1. Eigibility criteria
2.4. Data management
This bibliometric review focuses on recommender systems and AI
techniques in e-commerce. To select sources, we look for records with This bibliometric study used Microsoft Excel® for Microsoft 365
titles and keywords related to the subject. Second, records are included MSO (version 2403 build 16.0.17425.20124) 64-bit to extract, store,
that address the intersection of recommender systems and AI techniques and process information from selected databases. Microsoft Excel®
with e-commerce. This inclusive approach ensures a complete and ac­ made it easy to manage and analyze the data. The free program VOS­
curate representation of relevant research in this area. viewer® in its latest version 1.6.20 for Microsoft Windows Systems,
The exclusion process has three steps. The first is to remove records updated on October 31, 2023, was also used. This tool let us analyze how
with indexing errors. The second is to remove documents with missing terms appear together and create graphics showing this. This is the same
text. This only applies to systematic literature reviews. The third is to approach used in previous research (Kuzior & Sira, 2022). Microsoft
remove incomplete records. Finally, the third phase will focus on Excel® and VOSviewer® helped us understand patterns and

3
A. Valencia-Arias et al. Intelligent Systems with Applications 24 (2024) 200435

relationships in the scientific literature. contributed to the accurate and quantitative visualization of the results.
Microsoft Excel® is a good tool for data visualization. It has been
discussed in the literature (Raubenheimer, 2017) and is good for data 2.11. Assessment of reporting bias
representation and analysis, particularly in the context of bibliometrics.
Kuzior and Sira (2022) used VOSviewer® to analyze research on This paper looks at how a lack of results in the synthesis could lead to
blockchain technology. This supports using it to represent relationships bias. This bias could be due to reporting bias. Some synonyms in the
and patterns in the scientific literature related to recommender systems IEEE Thesaurus might affect the study. Also, texts with incomplete
and AI techniques in e-commerce. indexing may omit valuable information. These potential biases are
considered when interpreting and applying bibliometric results. They
2.5. Selection process affect how representative and complete the studies analyzed are.

Page et al. (2021) said it’s important to use internal automated 2.12. Assessment of certainty
classifiers to find relevant studies and to check that the studies are being
classified correctly. This study used Microsoft Excel® to do this. The This review looks at how sure we can be about the evidence. This
study investigators developed these tools and each team member used review looks at the whole study, not just one part. It looks at things like
them to assess inclusion and exclusion criteria. This approach was used how to choose which studies to include and exclude, how to measure
to ensure that no relevant studies were overlooked or misclassified. By things like how many times a study has been published, and how to spot
combining and validating the results, the selection process was made any problems with the study. These are explained in the method. The
more accurate and consistent. review also looks at how sure we can be about the results, points out any
problems with the study, and explains why the study is important
2.6. Data collection process (Fig. 1).
This bibliometric study involved selecting and excluding records.
The Page et al. (2021) guidelines say that we must describe how we First, records were found in each source, then duplicates were removed.
collected data from bibliographic reports. This means saying how many Then, three exclusion stages were applied, reflecting a comprehensive
reviewers worked on each report, whether they worked alone, and how screening process. After these stages, 91 articles were included. The
we checked the data with the study investigators. This bibliometric articles in this study are the basis for a full examination of recommender
research looked at recommender systems and AI techniques in e-com­ systems and AI techniques in e-commerce. This study’s methodology is
merce. Microsoft Excel was used to get data from the two databases. All based on a systematic and rigorous approach to data collection and
authors checked the data and did their work separately. They checked analysis.
the data together to make sure it was right. The results were checked
together to make sure they were reliable and consistent. 3. Results

2.7. Data elements This study shows important research trends. This figure shows a big
increase in scientific production, reflecting growing interest in the
This study collected results from articles that matched the research relationship between recommender systems and artificial intelligence
goal. We used search equations in each database to find these articles. If strategies in the context of electronic commerce. The most articles on
we found unclear or incomplete information, we excluded it. This was to this topic were published in 2020, 2021, 2022, and 2023. This is an
keep the results consistent and to ensure the integrity of the results. excellent trend in recent research.
The analysis of the main authors in recommender systems and AI
2.8. Assessment of the risk of bias of the study techniques in e-commerce shows different approaches and research
strategies. It shows that being productive is important and that you can
This study collected results from articles that matched the research make a big impact if you focus your work. Two groups were identified,
goal. We used search equations in each database to find these articles. If as shown in Fig. 3. The first group includes Paraschakis and Nilsson, who
we found unclear or incomplete information, we excluded it. This was to have made a big impact with their work. These studies have made a big
keep the results consistent and to ensure the integrity of the results. impact on the literature. The second group of authors is distinguished by
a high impact in terms of scientific productivity, although they have a
2.9. Impact measures relatively low index. Authors like Zhou L., Friedman D., Tan F., Zhang
Y., and Zhao Q. have made important contributions to research.
The specification of effect measures is distinct in this research. It is Fig. 4 shows that the most important journals in terms of citations
based on secondary sources such as scientific articles and metadata. can be divided into three groups. The first group includes journals like
Traditional impact measures are not applied in the same manner. Electronic Commerce Research, which are very productive and im­
Instead, bibliometric indicators are used. The results are presented in pactful. The second group includes journals with low scientific pro­
terms of the number of publications, the number of citations, and the ductivity but high impact. Examples include Recsys 2015-Proceedings
pattern of temporal use of each keyword. Furthermore, VOSviewer® is Of The 9th Acm Conference On Recommender Systems and Journal Of
used to analyze existing nodes and identify thematic associations. Visual Communica-tion And Image Representation. The third group is
distinguished by high scientific productivity, but less impact in terms of
2.10. Synthesis methods citations. Applied Soft Computing is a good example of a journal in this
category.
This process explains how studies are chosen for the synthesis. It also A detailed analysis of countries shows three groups (Fig. 5). The first
explains how the data is prepared and the results are presented. For this, group includes China, India, and the United States, which are major
bibliometric indicators were used to assess quantity, quality, and players in science. The second group includes countries like Sweden and
structure. These indicators follow the guidelines for scientific publica­ Spain, which have a moderate scientific productivity index but a high
tions (Durieux & Gevenois, 2010). These indicators were automatically impact on the subject. This shows that quality is more important than
applied to documents that passed the three exclusion stages. This quantity. The third group is distinguished by high scientific productiv­
approach allowed for an unbiased assessment of the quantity and quality ity, but not necessarily by the number of citations.
of publications, as well as the identification of structural patterns. This Fig. 6 shows how the literature on recommender systems and AI

4
A. Valencia-Arias et al. Intelligent Systems with Applications 24 (2024) 200435

Fig. 1. PRISMA flowchart. Own elaboration from Scopus and Web of Science.

techniques in e-commerce has changed over time. This analysis shows analysis and collaborative filtering.
the most common keywords in each year of the study, from 2007 to
2023. The year 2007 marks the start of the research, when multiagent 4. Discussion
systems emerged. In recent years, there has been a lot of research on
topics like sentiment analysis, convolutional neural networks, collabo­ The results are based on a methodical approach, following the
rative filtering, session-based recommendation, and implicit feedback. PRISMA-2020 guidelines. This methodology provides a structured
This research provides a more complete picture of the field. framework for a comprehensive bibliometric study, allowing the iden­
Fig. 7 shows the main keyword co-occurrence network, which has six tification and synthesis of relevant information in a transparent and
thematic clusters. These are shown in different colors, with yellow being reproducible manner. The methodology ensures that the data is
the most prominent. This cluster includes terms like collaborative searched, selected, and analyzed correctly, which makes the results
filtering, sparse rating, implicit feedback, and recommendation, which more valid and reliable.
show its importance in the field. The red cluster is about sentiment The discussion section of this report explains the results in detail. The
analysis, neural networks, accuracy, and Bert. Other clusters can be findings are presented in detail, allowing a deep understanding of the
identified in light blue, purple, green, and blue. These demonstrate how trends, how keywords interact, and how the field of recommender sys­
the keywords are related to recommendation systems and artificial in­ tems and artificial intelligence techniques in electronic commerce has
telligence techniques in electronic commerce. evolved. Also, we discuss the practical results and the study’s limita­
Fig. 8 shows how keywords are used in recommender systems and e- tions. Also, the keywords are grouped by function, the biggest gaps in
commerce. This Cartesian plane has two axes. The X-axis shows how current research are found, and a plan for future research is made.
often keywords are used, and the Y-axis shows how useful they are over
time. This makes four different quadrants.
In Quadrant 2 are terms that are rarely used but are important. These 4.1. Analysis of the growth of scientific literature on recommender
include terms like content-based image retrieval, content-based systems and ai techniques in e-commerce
filtering, convolutional neural network, and knowledge graph. In
Quadrant 1, we find consolidated and emerging concepts. These are A lot of scientific research was done in the fields of recommender
used a lot and are always interesting. Examples include sentiment systems and artificial intelligence techniques applied to e-commerce
between 2020 and 2023 (see Fig. 2). In 2020, a study showed how to

5
A. Valencia-Arias et al. Intelligent Systems with Applications 24 (2024) 200435

Fig. 2. Publications per year. Compiled from Scopus and Web of Science.

Fig. 3. Main authors. Compiled from Scopus and Web of Science.

make personalized recommendations in e-commerce using deep 4.2. Analysis of research references on recommender systems and AI
learning clustering (S. Wang & Qiu, 2021). In 2022, a study was done to techniques systems and AI techniques in e-commerce
deal with sparse score matrices in e-commerce recommender systems.
The study proposed a hybrid deep learning approach based on LSTM, Researchers Paraschakis D and Nilsson BJ have shown they are
SDAE, and latent factor (Pujastuti, Laksito, Hardi, Perwira & Arfriandi, leaders in recommender systems and AI techniques in e-commerce. In
2022). Another study from that year created a hierarchical e-commerce one of their papers, they compared the best recommender systems in an
recommendation system based on online users’ opinions (Islek & Ogu­ industrial context. This helped them understand how well recommender
ducu, 2022). systems work in e-commerce (Paraschakis, Nilsson & Holländer, 2015).
In 2021, a machine learning-based e-commerce recommendation A few years later, these authors conducted an experiment using
system was introduced (Tahir, Enam & Mustafa, 2021). A deep neural Thompson sampling to generate recommendations, presenting new ap­
network model was developed for e-commerce fashion recommenda­ proaches in e-commerce (Brodén, Hammar, Nilsson & Paraschakis,
tions using additional information (S. Wang & Qiu, 2021). Finally, in 2018).
2023, various authors developed a hybrid recommendation system for He also worked with researcher Zhou L on a recommendation system
e-commerce that focuses on the customer and uses sentiment analysis for for e-commerce platforms. This is an innovative way to improve
greater effectiveness (Karn et al., 2023). advertising recommendations (Zhou, 2020).
Researchers are interested in developing and improving recom­ Zhao Q and other researchers also looked at e-commerce recom­
mender systems and AI techniques for e-commerce. These systems mendations with personalized promotions. They came up with a way to
address challenges in an evolving field. combine product recommendations and promotion strategies for each
user. This approach improves user experience and sales in e-commerce
(Zhao, Zhang, Friedman & Tan, 2015). Together, these studies show
how authors influence recommender systems and AI techniques in

6
A. Valencia-Arias et al. Intelligent Systems with Applications 24 (2024) 200435

e-commerce. for e-commerce with Khan, Hussain, Iltaf, Kim and Jeon (2021). These
The following journals have helped improve research in recom­ countries have helped to improve knowledge and research in recom­
mendation systems and AI techniques applied to e-commerce: Electronic mender systems and AI techniques.
Commerce Research, Recsys 2015-Proceedings Of The 9th Acm Con­ It’s important to recognize the impact of contributions that aren’t
ference On Recommender Systems, Journal Of Visual Communication mentioned in the results but are important in the field. Song, Wu, Song
And Image Representation, and Applied Soft Computing. (See Fig. 4.) and Stojanovic (2023) study how to estimate states in reaction-diffusion
The journal Electronic Commerce Research has articles about adver­ neural networks under attack. This helps us understand how to keep
tising recommendations that help identify communities in complex neural systems safe. Similarly, Zhang, Song, Sun and Stojanovic (2023)
networks. These include Zhou (2020) and Chattopadhyay et al. (2021). present a hybrid approach based on fuzzy security for secure filtering in
The 2015 ACM Conference on Recommender Systems, called Recsys, nonlinear parabolic partial differential equation systems affected by
presents a method for recommending electronic commerce with cyber attacks.
personalized promotions. This expands the scope of recommendation Also, although they don’t focus on recommendation systems in e-
strategies (Zhao et al., 2015). commerce, works like Wei, Huang and Fu (2007) give a general over­
The Journal of Visual Communication and Image Representation is view of recommendation systems in e-commerce, while Schafer, Kon­
notable for its contributions to recommender systems and artificial in­ stan and Riedl (1999) present a pioneering perspective on
telligence techniques in electronic commerce. It analyzed personalized recommendation systems in e-commerce, establishing important foun­
recommendations in electronic commerce through clustering, which dations in the field. These contributions are essential to understanding
generated different perspectives on the effectiveness of this approach. e-commerce and recommender systems.
Similarly, Applied Soft Computing has a study on a contextual recom­ This literature review looks at how bibliometrics relate to recom­
mendation system for e-commerce. This study helps us understand how mendation systems and artificial intelligence in e-commerce. It also
to consider context in recommendations. Taken together, these journals looks at where different types of populations have been studied to un­
have helped develop knowledge in recommender systems and AI tech­ derstand the topic’s scientific distribution. Fig. 9 shows where this
niques in e-commerce. research was done.
China, India, and the United States have led the way in scientific The findings show that the theme has been most validated in Asia
research on recommender systems and AI techniques in e-commerce. and Europe. Many Asian countries, including Turkey, Israel, India,
China has done studies on hybrid recommendation systems for e-com­ China, South Korea, and Japan, have led the way in studying recom­
merce websites (Wenxing, Yang, Shunzhi & Maoqing, 2010) and deep mendation systems and artificial intelligence in the context of electronic
learning-based advertising recommendations (Zhou, 2020). India has commerce. Similarly, in Europe, countries like Sweden, Spain, Portugal,
also made contributions to this field. Patil and Rao (2019) explored the the Netherlands, Belgium, Germany, and Italy have also addressed this
role of AI in e-commerce website design, and Addagarla and Amalana­ issue. These are just a few examples. The research covers many more
than (2020) presented a research approach. In the United States, countries.
Agichtein and his team held a workshop at the WSDM 2020 conference
on conversational systems for e-commerce recommendations and
searches (Agichtein, Hakkani-Tür, Kallumadi & Malmasi, 2020). 4.3. Analysis of the thematic evolution of recommender systems and AI
Other countries have also made important contributions. In Sweden, techniques in e-commerce
Brodén et al. (2018) did a pilot study on recommendations in e-com­
merce. In Spain, Vanesa Aciar, Serarols-Tarres, Royo-Vela and Esteva In the early days of research on recommender systems and AI tech­
(2007) worked on improving e-commerce through recommendations niques in e-commerce, the idea of "multiagent systems" was a big part of
using intelligent agents. Indonesia has also made contributions to the the development (see Fig. 6). The work "Increasing effectiveness in e-
field, with research by Pujastuti et al. (2022) on recommender systems commerce: recommendations applying intelligent agents" was a key
for sparse evaluation matrices in e-commerce. South Korea has also been moment in understanding and using intelligent agents in e-commerce. It
a leader in this area, developing a contextual recommendation system looked at how collaborative and autonomous systems could help online
shops make better product and service recommendations (Vanesa Aciar

Fig. 4. Main journals. Compiled from Scopus and Web of Science.

7
A. Valencia-Arias et al. Intelligent Systems with Applications 24 (2024) 200435

Fig. 5. Main Countries. Own elaboration from Scopus and Web of Science.

Fig. 6. Thematic evolution. Compiled from Scopus and Web of Science.

et al., 2007). These early works laid the groundwork for future de­ Another important concept is "collaborative filtering." This was
velopments in the field. They expanded our understanding of topics like studied a lot in 2021. A machine learning system was created that uses
sentiment analysis, convolutional neural networks, collaborative collaborative filtering to make e-commerce recommendations more
filtering, session-based recommendation, and implicit feedback, which personal. This shows how users can help each other to make better
are now used in e-commerce. recommendations (Anitha & Kalaiarasu, 2021). In 2022, researchers
In 2019, we focused on "implicit feedback" because it helped us find focused on "convolutional neural networks" (CNNs). They developed a
ways to improve recommendations in e-commerce. In 2020, session- content-based recommendation system for online stores that used CNNs
based recommendations were important. This year, a framework for to improve recommendations. This research showed how computer
exploring and exploiting diverse recommendation components was vision can be used in this context (De, Banerjee, Rath, Swain & Samant,
explored, emphasizing adapting to changing user preferences in online 2022).
browsing sessions (Brodén, Hammar, Nilsson & Paraschakis, 2020). In e-commerce, several concepts have emerged as important areas of

8
A. Valencia-Arias et al. Intelligent Systems with Applications 24 (2024) 200435

Fig. 7. Keyword co-ocurrence network. Own elaboration from Scopus and Web of Science.

research. One of these is "sentiment analysis," which has become very (2015), who compared top-n recommenders from an industry perspec­
important in 2023. Since that year, researchers have studied how to tive. Hanafi (2022) suggested using different techniques to improve
combine different types of analysis to create better recommendations for e-commerce recommender systems. Mixing SDAE, attention mecha­
customers in e-commerce (Karn et al., 2023). nisms, and matrix factorization helps e-commerce recommender sys­
These concepts are an important step forward in the development of tems give more accurate ratings. Kottage, Jayathilake, Chankuma,
recommender systems and AI techniques in e-commerce. They offer new Ganegoda and Sandanayake (2018) developed a preference-based
ways to make online shopping better. recommendation system for clothing e-commerce sites.
The second most relevant cluster, in red, includes keywords like
"sentiment analysis," "neural network," "accuracy," and "Bert." This
4.4. Analysis of the thematic clusters on recommender systems and AI cluster shows how sentiment analysis and neural networks can improve
techniques in e-commerce e-commerce recommendations. It also shows how users’ feelings affect
personalised recommendations (Karn et al., 2023).
The main keyword co-occurrence network shows different groups of Thematic clusters show what researchers are interested in and how
related terms in the scientific literature on recommender systems and AI different ideas in recommender systems and AI techniques in e-com­
techniques in e-commerce. The yellow cluster is notable for its terms, merce connect.
including "collaborative filtering," "sparse rating," "implicit feedback,"
and "recommendation." This cluster is about improving e-commerce
recommendations. This is shown in the work of Paraschakis et al.

Fig. 8. Validity and frequency of keywords. Own elaboration from Scopus and Web of Science.

9
A. Valencia-Arias et al. Intelligent Systems with Applications 24 (2024) 200435

4.5. Analysis of the frequency and conceptual validity around the online experience (Anitha & Kalaiarasu, 2021). Other authors have
recommender systems and AI techniques in electronic commerce studied how collaborative recommendation approaches can be used in
the e-commerce real estate market.
The Cartesian plane in the results section shows how recommender
systems and AI techniques in electronic commerce are developing. 4.6. Classification of keywords related to recommender systems and AI
Keywords in this quadrant are "content-based image retrieval" and techniques in e-commerce according to their function
"content-based filtering" (see Fig. 8). These terms show that more and
more people are interested in using image-based content filtering and Table 1 shows the most common and fastest-changing keywords in
retrieval techniques in e-commerce. These concepts are becoming more recommender systems and AI techniques in e-commerce. They are
important because users want personalized recommendations that look grouped by their function. This classification helps us understand the
good. main characteristics and uses of each category, giving us a good over­
Content-based image retrieval helps e-commerce sites make better view of the latest developments in this field.
recommendations by combining visual and content information about
products. Images are important for choosing products and having a good
4.7. Role and literature of optimization algorithms
experience (Bagwari, Sinha, Singh, Garg & Kanti, 2022). Content-based
filtering looks at the content of items to generate personalized recom­
AI has helped e-commerce by improving recommendations and
mendations. In e-commerce, it helps make recommendations more
decision-making. These algorithms, inspired by biology and nature,
relevant by looking closely at the content of products (Tantyo &
have shown they can solve complex problems and optimize systems in
Mauritsius, 2022). These new ideas could help improve accuracy and
personalization in e-commerce recommender systems.
Quadrant 1 shows the concepts that are growing fast and are at the Table 1
Classification of keywords according to their function. Own elaboration from
forefront of e-commerce recommender system research and AI tech­
Scopus and Web of Science.
niques. These include "sentiment analysis" and "collaborative filtering".
These concepts are still important because they help make e-commerce Keyword Associated Tools Applications Characteristics
recommendations more personal and accurate. Content-Based Visual Feature Product Use visual
Sentiment analysis helps businesses understand what customers Image Extraction, Recommendation features
think about their products and services. This helps them make better Retrieval Image Analysis
Content-Based User Profiling, Personalized Content Based on user
recommendations. Some examples of this are customer-centric hybrid Filtering Text Analysis Recommendation preferences
recommendation systems that integrate sentiment analysis. These have Convolutional Deep Learning, Image Recognition, Image
been shown to improve the user experience in e-commerce. On the other Neural Image Content Analysis recognition and
hand, problems and solutions in recommendation systems enabled by Network Processing analysis
Knowledge Graph Database, Knowledge Structured
deep learning have also been explored, showing how sentiment analysis
Graph Semantic Representation, knowledge
helps optimize such systems (Almahmood & Tekerek, 2022). Modeling Search representation
Collaborative filtering is a reliable way to make recommendations. Sentiment NLP, Machine Customer Review Evaluate
This approach uses past user-product interactions to provide relevant, Analysis Learning Analysis, Feedback emotions in text
personalized recommendations. This method improves recommenda­ Collaborative User-Item Recommendation, Based on user
Filtering Interaction, Group Decision- behavior
tions by using information on user behavior. Some authors have created Matrix Making
recommender systems based on machine learning that use collaborative Factorization
filtering to improve recommendations. This helps users and improves

Fig. 9. Global distribution of the research. Own elaboration from Scopus and Web of Science.

10
A. Valencia-Arias et al. Intelligent Systems with Applications 24 (2024) 200435

changing environments. One of the most important methods is the Recommendation systems affect how consumers buy things. It’s
gravitational search algorithm, which shows how celestial bodies move important to make these systems fair and honest. Your study shows that
under the influence of gravity. This method has been used to improve people are using more complex techniques like sentiment analysis and
recommendation systems, making them more accurate and efficient collaborative filtering. This means we need to understand and deal with
(Kourepinis, Iliopoulou, Tassopoulos & Beligiannis, 2024; Somu, MR, any bias or discrimination in the recommendations. Algorithms should
Kaveri, Krithivasan & VS, 2020). be transparent, understandable, and fair to users. They should also
The literature shows that optimizing the system of inclined planes is provide control and explanation of how recommendations are
a promising strategy in e-commerce. This approach has been used to generated.
improve pricing and online purchases. These methods help solve com­ New techniques like "session-based recommendation" and "implicit
plex e-commerce problems. They use natural principles and algorithms feedback" show that companies are trying to make online recommen­
to improve recommendation systems (Mozaffari, Abdy & Zahiri, 2016; dations more personal. This can help customers find what they want
Sarkar et al., 2023). more easily and make them happier. This trend also shows the impor­
The latest research shows that these methods can improve e-com­ tance of having a technology that can process and respond to user in­
merce profits and efficiency. For example, Mozaffari et al. (2016) show teractions in real time. This helps to create a personalized experience.
how the inclined plane optimization (IPO) system can help solve opti­ This bibliometric analysis also shows ways that recommender sys­
mization problems in many areas, including e-commerce. Khamees, tems in e-commerce can be improved. New techniques like image
El-Rafei, Badra and Abdelaziz (2017) also look at how evolutionary al­ retrieval and content-based filtering suggest a promising approach to
gorithms can help solve problems in electronic commerce. recommendation personalization. This goes beyond considering text.
Integrating images and other content into recommendation algorithms
4.8. Practical implications can enhance the user experience and improve recommendations. This is
especially useful in sectors like fashion, interior design, and gastronomy,
This analysis offers insights and guidance for researchers, pro­ where appearance and style are important.
fessionals, and companies in this evolving field. The shift from studying The study shows that it is important to work together in different
"Multiagent Systems" to analyzing "Sentiment Analysis," "Convolutional fields to create better recommendation systems. The combination of
Neural Networks," "Collaborative Filtering," "Session-Based Recom­ sentiment analysis, artificial intelligence, and consumer psychology can
mendation," and "Implicit Feedback" reflects a change in how re­ help us understand how people think and act. There has been little
searchers approach these topics, which have been adapted to the needs research on how to use psychological data in these algorithms, such as
of the current market. This shift suggests that researchers and practi­ how consumers see themselves (Feng, 2023). This can help companies
tioners should consider new techniques to improve e-commerce understand what customers want and how they act in online stores.
recommender systems. Companies can work with experts to create better recommendation
The main topic cluster around terms like "collaborative filtering," systems. These systems should consider what customers say and what
"sparse rating," "implicit feedback," and "recommendation" shows how they really want.
these concepts are connected. This provides a good path for further
research and development. This grouping suggests that focusing on user 4.9. The practical applications of this theory in real-world settings
collaboration, handling implicit feedback, and improving recommen­
dations is important. E-commerce organizations and professionals can Two new techniques let websites change recommendations based on
benefit from using these strategies to improve the user experience and what users do online. This is especially useful in e-commerce, where
customer loyalty. suggestions can affect purchases. For example, if someone searches for
The keyword analysis shows that new ideas are becoming more summer clothes, they might get suggestions for similar items based on
popular. These include "Content-Based Image Retrieval," "Convolutional what they’ve been looking at. This makes the customer experience
Neural Network," and "Knowledge Graph." This shows that image better and increases the chance of them buying.
analysis and knowledge graphs are becoming more important. Ecom­ Integrating sentiment analysis with collaborative filtering helps
merce companies can use these new approaches to give customers bet­ businesses understand customers better, which is good for personalizing
ter, more personalised recommendations. This will make customers shopping experiences and making better marketing decisions. An online
happier and more likely to buy. At the same time, more focus on store can analyze product reviews to identify positive or negative sen­
sentiment analysis and collaborative filtering shows how important it is timents and adjust its recommendations and promotions. This helps
to understand what customers want and how they feel. This will help companies understand what customers want and need, which makes
companies market their products better and make them more customers happier and more loyal.
personalised. This is useful in fashion, interior design, and gastronomy, where
Because customer preferences and market trends change all the time, images are important. A fashion app could let users upload images of
ecommerce companies need to keep learning and adapting their clothes they like and get recommendations of similar items in the store.
recommendation systems. This study shows that keeping recommen­ Adding visual content to recommendation systems makes suggestions
dation models up to date is important. Professionals and companies more accurate and makes shopping easier for users.
should think about ways to get new data and adapt models to detect Recommender systems can use knowledge graphs to suggest prod­
changes in user behavior and product offerings. Updating approaches ucts based on more than just what a user has bought. For example, an
can help keep recommendations relevant and accurate over time. online bookstore could suggest books based on topics, favorite authors,
Artificial intelligence is changing how businesses and consumers buy and reviews from other readers with similar interests. This makes rec­
things online. This means we must think about the challenges and op­ ommendations more personal and relevant, which can increase sales and
portunities in terms of technology, money, and ethics. Academics also customer satisfaction.
face new challenges in this new paradigm (Krishna et al., 2023).
Personalized recommendation systems are also affecting e-commerce. 4.10. Limitations
Molaie and Lee (2022) say these systems can affect sales diversity.
Conversely, content filtering can expose users to niche products and The current research on recommender systems and AI techniques in
make sales more even. This shows how important it is to design e-commerce has some limitations. First, the databases used, like Scopus
recommendation systems to influence sales and the e-commerce and Web of Science, might have left out other useful sources of infor­
economy. mation, which could have made the data incomplete. Also, the search

11
A. Valencia-Arias et al. Intelligent Systems with Applications 24 (2024) 200435

terms could have missed some important contributions because the recommender systems and AI techniques in e-commerce.
quality and precision of the search terms could have been poor.
The current bibliometric analysis has focused on two databases, 4.12. Research agenda
Scopus and Web of Science. This may have left out other sources of in­
formation. The two databases may not have included all research on This study proposes a research agenda for AI recommender systems
recommender systems and AI in e-commerce. Future research should in e-commerce. It includes the main, growing, and emerging keywords
include more databases. To get a fuller picture of the research in this from the scientific literature on the topic. Fig. 10 shows the current state
fast-changing field, it’s best to include both academic and industrial of knowledge in this area.
sources. This could reveal new trends, approaches, and contributions The tool lets you study patterns, users, trends, and hidden relation­
that have not been captured, enriching global understanding and ships in large amounts of data. Future research should study how to use
providing a solid foundation for decision-making in research and advanced data mining to better understand how users interact with e-
development in the field of electronic commerce. commerce platforms. Also, combining data mining with other AI
Also, when using Excel and VOSviewer to analyze bibliometric data, methods, like neural networks and deep learning, could create more
there is a chance of human error. Even though we try to avoid mistakes, intelligent and personalized recommendation systems. It would also be
computers can make them. The results could be biased because re­ good to look at how data mining can help to stop people from cheating
searchers decide what to include and how to interpret the data. and to make online shopping safer.
Another limitation is that the research field of recommender systems Sentiment analysis is a key tool in e-commerce recommendation
and AI techniques in e-commerce is changing and dynamic. Some recent systems and AI techniques. This technique is used to identify user
developments and advances may not have been captured in the data, preferences and to detect satisfaction or dissatisfaction with e-commerce
which could affect the results. The present bibliometric analysis must be platforms. Future studies should look at how to improve sentiment
interpreted with caution because of its limitations. This is to obtain a analysis and how it can be used in recommendation algorithms to make
more complete and accurate understanding of the research landscape in more intelligent systems that take into account the emotional side of
this field. purchases. It would also be useful to see if sentiment analysis can be used
Also, the research doesn’t have a more detailed statistical analysis or in different cultures and languages to make recommendations that work
mathematical modeling to show trends and correlations. While tools like for everyone. Finally, we should look into how sentiment analysis can
Microsoft Excel® and VOSviewer® were used to show how the data is help users and e-commerce platforms interact better through chatbots
related, they don’t show how the data changes over time. The research that understand customers.
can’t identify patterns and relationships between the variables if there’s The field of recommender systems and AI techniques in e-commerce
no more rigorous statistical analysis. Future studies should use more has been the subject of interest in recent times due to the potential of
advanced statistical techniques to understand the research landscape in Recurrent Neural Networks (RNNs) to model data streams. Currently,
this field better. RNNs are being used to predict purchase patterns and user preferences
The present bibliometric analysis covers a wide time period without based on their history, which enables the selection of more relevant and
focusing on the most recent developments. Future studies could benefit timely recommendations. Future research could look at ways to make
from a more rigorous investigation of recommender systems and artifi­ RNNs faster and more efficient, especially in e-commerce. New RNN
cial intelligence techniques. The last five years of literature would show architectures like LSTM and GRU could help with long-term temporal
the most current and emerging trends. This would provide a more up-to- dependencies. It would also be useful to see how RNNs could be used to
date and relevant perspective on recent advances and innovations in create product descriptions based on user behavior to improve online
electronic commerce. Also, focusing on the recent period helps identify shopping.
technologies and methods that are becoming more common in the in­ Collaborative filtering is still important in e-commerce. This is based
dustry. This would help future research and development. on the idea that similar users’ preferences and behaviours can be used to
make recommendations to a particular user. Currently, collaborative
4.11. Research gaps filtering is being improved by new approaches based on matrices and
factorization models. However, future research could make this tech­
Table 2 lists the main problems in recommender systems and AI nique better by finding new ways to deal with the "cold start" problem
techniques in e-commerce. These are important areas for future and data in large systems.
research. These deficiencies result from an analysis of the literature and The rise of big data has changed how e-commerce uses recommender
show where current research has not yet provided satisfactory answers systems and AI. Currently, big data is used to track customer interactions
or where there are opportunities for advancing knowledge. The table in real time, allowing recommendations to be changed quickly. Future
shows what needs to be done and why. This approach helps researchers studies could improve this by looking at how to use big data tools like
find promising areas of study and a plan for future research in Apache Hadoop and Spark to make recommender systems faster and

Table 2
Research gaps. Compiled from Scopus and Web of Science.
Category Investigative Gaps Justification Future Research Questions

Thematic Gaps 1. Use AI and ethics in recommendations. 2. Ethics in AI is important because it affects society How can ethical principles be added to
How AI is used in e-commerce. and the economy. New e-commerce techniques recommendation algorithms? How can AI improve e-
need to be studied. commerce personalization?
Geographic gaps 1. Research on recommender systems in Most research is on developed countries, so we How can recommender systems be adapted to emerging
emerging markets. 2. Research on electronic don’t know much about developing markets (see markets? How can we work together to improve e-
commerce between countries. Fig. 9). International collaboration can help. commerce research?
Interdisciplinary 1. Personalize recommendations with Personalization could benefit from psychology. How can consumer decision-making theories help with
Gaps psychological approaches. 2. Marketing Marketing can make recommendations more personalization? How can traditional marketing help
methods for recommendations. effective. improve user satisfaction with recommendation
systems?
Temporary Gaps 1. How have recommender systems been Long-term research can show how things change How have recommender systems changed over time?
used in e-commerce? over time. AI is changing fast. We need to look What AI techniques will impact e-commerce in the
2. Study of how AI will affect e-commerce. ahead. future?

12
A. Valencia-Arias et al. Intelligent Systems with Applications 24 (2024) 200435

Fig. 10. Research agenda. Own elaboration from Scopus and Web of Science.

more efficient. Also, studying more advanced analytical techniques like The authors Paraschakis D and Nilsson BJ, the journal Electronic
stream mining and graph analysis could help identify more useful pat­ Commerce Research, and countries such as China, India, and the United
terns in e-commerce data. States have been identified as the primary research references. This
Convolutional neural networks (CNNs) are important in recom­ overview shows the main people and sources in research on recom­
mender systems and AI techniques in e-commerce because they can mender systems and AI in e-commerce.
process and extract features from unstructured data. Currently, CNNs The focus of scientific research has changed a lot. At first, researchers
are being used to improve recommendations by incorporating images focused on Multiagent Systems. Now, they study concepts like Sentiment
and user feedback. Future research could explore how to adapt and Analysis, Convolutional Neural Networks, Collaborative Filtering,
optimize CNNs for real-time e-commerce data processing. Also, Session-Based Recommendation, and Implicit Feedback. The research
combining CNNs with other recommendation methods could create community can adapt its approaches to new technology and e-commerce
more powerful and versatile hybrid systems. It would also be good to needs. The analysis shows that keywords are related. The cluster formed
look at how CNNs can automatically identify features in product images, by "Collaborative Filtering," "Sparse Rating," "Implicit Feedback," and
which would help make products more personalised and attractive. "Recommendation" is particularly strong. This shows that the research
Knowledge graphs are important in e-commerce and AI. They are areas are connected.
used to make recommendations more relevant by considering relation­ Keywords related to emerging and growing topics include sentiment
ships between products and users. Future research could look into analysis and collaborative filtering. Other terms to watch are content-
building and maintaining knowledge graphs in real time. Also, we could based image retrieval, content-based filtering, convolutional neural
combine knowledge graphs with other recommendation techniques to network, and knowledge graph. These topics offer promising research
create more informed and contextual systems. avenues for understanding how new trends influence recommender
systems in e-commerce.
5. Conclusions Finally, this study has looked at how recommender systems and AI
techniques in e-commerce have changed over time. The results can help
This bibliometric study found that 2021 and 2022 were the most future researchers understand more about the issues in this field and
interesting years for this subject. This shows that research in this field is how to improve it.
constantly changing and growing.

13
A. Valencia-Arias et al. Intelligent Systems with Applications 24 (2024) 200435

Author contributions and research agenda. Frontiers in Artificial Intelligence, 6, Article 1167735. https://
doi.org/10.3389/frai.2023.1167735
Hanafi, H. (2022). Enhance rating prediction for e-commerce recommender system using
Removed for peer review. hybridization SDAE, attention mechanism and matrix factorization. https://doi.org/
10.22266/ijies2022.1031.37
He, X., & Liu, Y. (2024). Knowledge evolutionary process of Artificial intelligence in E-
Funding commerce: Main path analysis and science mapping analysis. Expert Systems with
Applications, 238, Article 121801. https://doi.org/10.1016/j.eswa.2023.121801
This research did not receive any specific grant from any funding Islek, I., & Oguducu, S. G. (2022). A hierarchical recommendation system for E-
commerce using online user reviews. Electronic Commerce Research and Applications,
agency in the public, commercial, or nonprofit sectors. 52. https://doi.org/10.1016/j.elerap.2022.101131
Karn, A. L., Karna, R. K., Kondamudi, B. R., Bagale, G., Pustokhin, D. A., Pustokhina, I. V.,
et al. (2023). Customer centric hybrid recommendation system for E-Commerce
applications by integrating hybrid sentiment analysis. Electronic Commerce Research,
Declaration of competing interest
23(1), 279–314. https://doi.org/10.1007/s10660-022-09630-z
Khamees, A. K., El-Rafei, A., Badra, N. M., & Abdelaziz, A. Y. (2017). Solution of optimal
The authors declare that they have no known competing financial power flow using evolutionary-based algorithms. International Journal of Engineering,
Science and Technology, 9(1), 55–68. https://doi.org/10.4314/ijest.v9i1.5
interests or personal relationships that could have appeared to influence
Khan, Z., Hussain, M. I., Iltaf, N., Kim, J., & Jeon, M. (2021). Contextual recommender
the work reported in this paper. system for E-commerce applications. Applied Soft Computing, 109, Article 107552.
https://doi.org/10.1016/j.asoc.2021.107552
Data availability Kottage, G. N., Jayathilake, D. K., Chankuma, K. C., Ganegoda, G. U., & Sandanayake, T.
(2018). Preference based recommendation system for apparel e-commerce sites. In
2018 IEEE/ACIS 17th international conference on computer and information science
Data will be made available on request. (ICIS (pp. 122–127). https://doi.org/10.1109/ICIS.2018.8466382
Kourepinis, V., Iliopoulou, C., Tassopoulos, I., & Beligiannis, G. (2024). An artificial fish
swarm optimization algorithm for the urban transit routing problem. Applied Soft
Computing, 155, Article 111446. https://doi.org/10.1016/j.asoc.2024.111446
References Krishna, I. M., Avinash, C. S. N., Jaganadham, R., Durga, A. J. L., Madhulika, Y., &
Rakesh, K. V. V. S. (2023). The impact of artificial intelligence effect on e-commerce:
A framework for key research areas. In In 2023 IEEE International Conference on ICT
Acharya, N., Sassenberg, A. M., & Soar, J. (2023). Effects of cognitive absorption on
in Business Industry & Government (ICTBIG) (pp. 1–6). IEEE.
continuous use intention of AI-driven recommender systems in e-commerce.
Kuzior, A., & Sira, M. (2022). A bibliometric analysis of blockchain technology research
Foresight (Los Angeles, Calif.), 25(2), 194–208. https://doi.org/10.1108/FS-10-2021-
using VOSviewer. Sustainability, 14(13). https://doi.org/10.3390/su14138206
0200
Langs, O. E. V. (2012). Enfoques no estándar de algoritmos evolutivos en un dilema de
Addagarla, S. K., & Amalanathan, A. (2020). Probabilistic unsupervised machine learning
optimización. Revista Mutis, 2(2), 126–138. https://orcid.org/0000-0002-0170-
approach for a similar image recommender system for E-commerce. Symmetry, 12
462X.
(11). https://doi.org/10.3390/sym12111783
Loukili, M., Messaoudi, F., & El Ghazi, M. (2023). Machine learning based recommender
Agichtein, E., Hakkani-Tür, D., Kallumadi, S., & Malmasi, S. (2020). ConvERSe’20: The
system for e-commerce. IAES International Journal of Artificial Intelligence, 12(4),
WSDM 2020 workshop on conversational systems for e-commerce recommendations
1803–1811. https://doi.org/10.11591/ijai.v12.i4.pp1803-1811
and search. In Proceedings of the 13th International Conference on Web Search and Data
Mohammadi, A., & Zahiri, S. H. (2018). Inclined planes system optimization algorithm
Mining (pp. 897–898). https://doi.org/10.1145/3336191.3371882
for IIR system identification. International Journal of Machine Learning and
Almahmood, R. J. K., & Tekerek, A. (2022). Issues and solutions in deep learning-enabled
Cybernetics, 9, 541–558. https://doi.org/10.1007/s13042-016-0588-x
recommendation systems within the e-commerce field. Applied Sciences, 12(21),
Molaie, M. M., & Lee, W. (2022). Economic corollaries of personalized recommendations.
11256. https://doi.org/10.3390/app122111256
Journal of Retailing and Consumer Services, 68, Article 103003.
Almutairi, Y., & Abdullah, M. (2020). IRHM: Inclusive review helpfulness model for
Mozaffari, M. H., Abdy, H., & Zahiri, S. H. (2016). IPO: An inclined planes system
review helpfulness prediction in e-commerce platform. Journal of Information
optimization algorithm. Computing and Informatics, 35(1), 222–240.
Technology Management, 12(2), 184–197. https://doi.org/10.22059/
Necula, S. C., & Păvăloaia, V. D. (2023). AI-Driven Recommendations: A Systematic
jitm.2020.75800
review of the state of the art in E-Commerce. Applied Sciences, 13(9), 5531. https://
Anitha, J., & Kalaiarasu, M. (2021). Optimized machine learning based collaborative
doi.org/10.1016/j.jretconser.2022.103003, 10.3390/app13095531.
filtering (OMLCF) recommendation system in e-commerce. Journal of Ambient
Page, M. J., McKenzie, J. E., Bossuyt, P. M., Boutron, I., Hoffmann, T. C., Mulrow, C. D.,
Intelligence and Humanized Computing, 12, 6387–6398. https://doi.org/10.1007/
et al. (2021). The PRISMA 2020 statement: An updated guideline for reporting
s12652-020-02234-1
systematic reviews. International Journal of Surgery, 88, Article 105906. https://doi.
Bagwari, A., Sinha, A., Singh, N. K., Garg, N., & Kanti, J. (2022). CBIR-DSS: Business
org/10.1016/j.ijsu.2021.105906
decision oriented content-based recommendation model for e-commerce.
Paraschakis, D., Nilsson, B. J., & Holländer, J. (2015). Comparative evaluation of top-n
Information, 13(10), 479. https://doi.org/10.3390/info13100479
recommenders in e-commerce: An industrial perspective. In 2015 IEEE 14th
Bawack, R. E., Wamba, S. F., Carillo, K. D. A., & Akter, S. (2022). Artificial intelligence in
International Conference on Machine Learning and Applications (ICMLA (pp.
e-commerce: A bibliometric study and literature review. Electronic Markets, 32(1),
1024–1031). https://doi.org/10.1109/ICMLA.2015.183
297–338. https://doi.org/10.1007/s12525-022-00537-z
Patil, M., & Rao, M. (2019). Studying the contribution of machine learning and artificial
Brodén, B., Hammar, M., Nilsson, B. J., & Paraschakis, D. (2018). Ensemble
intelligence in the interface design of e-commerce site. Smart Intelligent Computing
recommendations via thompson sampling: An experimental study within e-
and Applications: Proceedings of the Second International Conference on SCI 2018, 2,
commerce. In 23rd international conference on intelligent user interfaces (pp. 19–29).
197–206. https://doi.org/10.1007/978-981-13-1927-3_20
https://doi.org/10.1145/3172944.3172967
Pujastuti, E., Laksito, A., Hardi, R., Perwira, R., & Arfriandi, A. (2022). Handling sparse
Brodén, B., Hammar, M., Nilsson, B. J., & Paraschakis, D. (2020). A bandit-based
rating matrix for e-commerce recommender system using hybrid deep learning based
ensemble framework for exploration/exploitation of diverse recommendation
on LSTM. SDAE and Latent Factor. International Journal of Intelligent Engineering &
components: An experimental study within e-commerce. ACM Transactions on
Systems, 15(2), 10 22266 2022 0430 35.
Interactive Intelligent Systems (TiiS, 10(1), 1–32. https://doi.org/10.1145/3237187
Raubenheimer, J. (2017). Excel-lence in data visualization?: The use of microsoft excel
Cabrera-Sánchez, J. P., Ramos-de-Luna, I., Carvajal-Trujillo, E., & Villarejo-Ramos, Á. F.
for data visualization and the analysis of big data. Data visualization and statistical
(2020). Online recommendation systems: Factors influencing use in e-commerce.
literacy for open and big data (pp. 153–193).
Sustainability, 12(21), 8888. https://doi.org/10.3390/su12218888
Reyes, N. S. (2023). Uso de la inteligencia artificial en la personalización de la
Chattopadhyay, S., Basu, T., Das, A. K., Ghosh, K., & Murthy, L. C. (2021). Towards
experiencia del usuario en plataformas digitales. Polo del Conocimiento, 8(6),
effective discovery of natural communities in complex networks and implications in
1190–1206.
e-commerce. Electronic Commerce Research, 21, 917–954. https://doi.org/10.1007/
Roy, D., & Dutta, M. (2022). A systematic review and research perspective on
s10660-019-09395-y
recommender systems. Journal of Big Data, 9(1), 59. https://doi.org/10.1186/
Darband, R. B., Mahboob, A. S., Shahbandegan, S., & Razavi, S. M. (2020). Multimodal
s40537-022-00592-5
optimization using inclined planes system optimization algorithm. In 2020 6th
Sahu, G., & Gaur, L. (2024). Decoding the recommender system: A comprehensive guide
Iranian conference on signal processing and intelligent systems (ICSPIS) (pp. 1–5). IEEE.
to explainable AI in e-commerce. Role of explainable artificial intelligence in E-
https://doi.org/10.1109/ICSPIS51611.2020.9349594.
Commerce (pp. 33–52). Cham: Springer Nature Switzerland. https://link.springer.
De, U. C., Banerjee, S., Rath, M. K., Swain, T., & Samant, T. (2022). Content based
com/chapter/10.1007/978-3-031-55615-9_3#citeas.
apparel recommendation for e-commerce stores. In 2022 3rd International Conference
Sarkar, M., Ayon, E. H., Mia, M. T., Ray, R. K., Chowdhury, M. S., Ghosh, B. P., et al.
for Emerging Technology (INCET (pp. 1–6). https://doi.org/10.1109/
(2023). Optimizing E-commerce profits: A comprehensive machine learning
INCET54531.2022.9824870
framework for dynamic pricing and predicting online purchases. Journal of Computer
Durieux, V., & Gevenois, P. A. (2010). Bibliometric indicators: Quality measurements of
Science and Technology Studies, 5(4), 186–193. https://doi.org/10.32996/
scientific publication. Radiology, 255(2), 342–351. https://doi.org/10.1148/
jcsts.2023.5.4.19
radiol.09090626
Schafer, J. B., Konstan, J., & Riedl, J. (1999). Recommender systems in e-commerce. In
Feng, Y. (2023). Enhancing e-commerce recommendation systems through approach of
Proceedings of the 1st ACM conference on Electronic commerce (pp. 158–166).
buyer’s self-construal: Necessity, theoretical ground, synthesis of a six-step model,

14
A. Valencia-Arias et al. Intelligent Systems with Applications 24 (2024) 200435

Sharma, S. N., & Sadagopan, P. (2022). Influence of conditional holoentropy-based World Wide Web Conference (pp. 398–403). https://doi.org/10.1145/
feature selection on automatic recommendation system in E-commerce sector. 3308560.3316596
Journal of King Saud University-Computer and Information Sciences, 34(8), 5564–5577. Wang, S., & Qiu, J. (2021). A deep neural network model for fashion collocation
https://doi.org/10.1016/j.jksuci.2020.12.022 recommendation using side information in e-commerce. Applied Soft Computing, 110,
Shin, K., Kwak, H., Kim, K. M., Kim, M., Park, Y. J., Jeong, J., et al. (2021). One4all user Article 107753. https://doi.org/10.1016/j.asoc.2021.107753
representation for recommender systems in e-commerce. arXiv preprint. https://doi. Wei, K., Huang, J., & Fu, S. (2007). A survey of e-commerce recommender systems. In
org/10.48550/arXiv.2106.00573. arXiv:2106.00573 2007 international conference on service systems and service management (pp. 1–5).
Somu, N., MR, G. R., Kaveri, A., Krithivasan, K., & VS, S. S. (2020). IBGSS: An improved IEEE. https://doi.org/10.1109/ICSSSM.2007.4280214.
binary gravitational search algorithm based search strategy for QoS and ranking Wenxing, H., Yang, W., Shunzhi, Z., & Maoqing, L. (2010). Hybrid recommender system
prediction in cloud environments. Applied Soft Computing, 88, Article 105945. for vertical-e-commerce website [J. System Engineering Theory & Practice, 5(1),
https://doi.org/10.1016/j.asoc.2019.105945 928–929. https://doi.org/10.1109/ISCID.2015.140
Song, X., Wu, N., Song, S., & Stojanovic, V. (2023). Switching-like event-triggered state Zhang, Q., Lu, J., & Jin, Y. (2021). Artificial intelligence in recommender systems.
estimation for reaction–diffusion neural networks against DoS attacks. Neural Complex & Intelligent Systems, 7(1), 439–457. https://doi.org/10.1007/s40747-020-
Processing Letters, 55(7), 8997–9018. https://doi.org/10.1007/978-3-031-55615-9_3 00212-w
Tahir, M., Enam, R. N., & Mustafa, S. M. N. (2021). E-commerce platform based on Zhang, Z., Song, X., Sun, X., & Stojanovic, V. (2023). Hybrid-driven-based fuzzy secure
machine learning recommendation system. In 2021 6th International Multi-Topic ICT filtering for nonlinear parabolic partial differential equation systems with cyber
Conference (IMTIC (pp. 1–4). https://doi.org/10.1109/IMTIC53841.2021.9719822 attacks. International Journal of Adaptive Control and Signal Processing, 37(2),
Tantyo, H., & Mauritsius, T. (2022). Implementation of recommendation system in e- 380–398. https://doi.org/10.1002/acs.3529
commerce using approximate nearest neighbor. Journal Of Theoretical And Applied Zhao, Q., Zhang, Y., Friedman, D., & Tan, F. (2015). E-commerce recommendation with
Information Technology, 100(ue 12), 3885–389330. personalized promotion. In Proceedings of the 9th ACM Conference on Recommender
Vanesa Aciar, S., Serarols-Tarres, C., Royo-Vela, M., & Esteva, J. L. (2007). Increasing Systems (pp. 219–226). https://doi.org/10.1145/2792838.2800178
effectiveness in e-commerce: Recommendations applying intelligent agents. Zhou, L. (2020). Product advertising recommendation in e-commerce based on deep
International Journal of Business and Systems Research, 1(1), 81–97. https://doi.org/ learning and distributed expression. Electronic Commerce Research, 20(2), 321–342.
10.1504/IJBSR.2007.014774 https://doi.org/10.1007/s10660-020-09411-6
Wang, H., Amagata, D., Maekawa, T., Hara, T., Niu, H., Yonekawa, K., et al. (2019). Zhu, J., & Liu, W. (2020). A tale of two databases: The use of Web of Science and Scopus
Preliminary investigation of alleviating user cold-start problem in e-commerce with in academic papers. Scientometrics, 123(1). https://doi.org/10.1007/s11192-020-
deep cross-domain recommender system. In Companion Proceedings of The 2019 03387-8

15