Received May 15, 2020, accepted June 1, 2020, date of publication June 16, 2020, date of current version

July 2, 2020.
Digital Object Identifier 10.1109/ACCESS.2020.3002803

A Systematic Study on the Recommender

Systems in the E-Commerce
PEGAH MALEKPOUR ALAMDARI1 , NIMA JAFARI NAVIMIPOUR 2 ,
MEHDI HOSSEINZADEH 3,4 , ALI ASGHAR SAFAEI5 , AND ASO DARWESH6
1 Department of Computer Engineering, Qeshm Branch, Islamic Azad University, Qeshm 7953163135, Iran
2 Future Technology Research Center, National Yunlin University of Science and Technology, Douliu 64002, Taiwan
3 Institute of Research and Development, Duy Tan University, Da Nang 550000, Vietnam
4 Health Management and Economics Research Center, Iran University of Medical Sciences, Tehran 1449614535, Iran
5 Department of Biomedical Informatics, Faculty of Medical Sciences, Tarbiat Modares University, Tehran 14115-111, Iran
6 Department of Information Technology, University of Human Development, Sulaimaniyah 0778-6, Iraq

Corresponding author: Mehdi Hosseinzadeh ([email protected])

ABSTRACT Electronic commerce or e-commerce includes the service and good exchange through elec-
tronic support like the Internet. It plays a crucial role in today’s business and users’ experience. Also,
e-commerce platforms produce a vast amount of information. So, Recommender Systems (RSs) are a solution
to overcome the information overload problem. They provide personalized recommendations to improve
user satisfaction. The present article illustrates a comprehensive and Systematic Literature Review (SLR)
regarding the papers published in the field of e-commerce recommender systems. We reviewed the selected
papers to identify the gaps and significant issues of the RSs’ traditional methods, which guide the researchers
to do future work. So, we provided the traditional techniques, challenges, and open issues concerning
traditional methods of the field of review based on the selected papers. This review includes five categories of
the RSs’ algorithms, including Content-Based Filtering (CBF), Collaborative Filtering (CF), Demographic-
Based Filtering (DBF), hybrid filtering, and Knowledge-Based Filtering (KBF). Also, the salient points of
each selected paper are briefly reported. The publication time of the selected papers ranged from 2008 to
2019. Also, we provided a comparison table of important issues of the selected papers as well as the tables
of advantages and disadvantages. Moreover, we provided a comparative table of metrics and review issues
for the selected papers. And finally, the conclusions can, to a great extent, provide valuable guidelines for
future studies.

INDEX TERMS Recommendation systems, electronic commerce, personalization, review.

I. INTRODUCTION place. Some services, such as payment, selling and buying

Electronic commerce (e-commerce) is concerned with such digital goods and physical products, ticket and hotel reserva-
diverse activities as marketing, servicing, developing, deliver- tions, online registration, e-learning [5], electronic customer
ing, selling, and expending for services and products by using relationship management [6], and many others are widely
online platforms [1], [2]. Today, e-commerce is an insepara- available with these applications. E-commerce websites, such
ble part of the business for a variety of reasons, including the as eBay, Alibaba, Amazon, and Netflix, provide great oppor-
ease of use, universal accessibility, wide variety, and manage- tunities for their users. These e-commerce platforms tend
able compassion of products from different vendors, trusted to utilize the social network features [7], expert cloud tech-
payment ways [3], and the convenience of home shopping nology [8]–[12], big data [13], and using mobile devices to
with the least waste of time. These facilities of e-commerce provide user satisfaction for their users.
applications improved the lives of users to a highlevel of The Recommender Systems (RSs) learn the experi-
quality, particularly with the technological development of ences and opinions from their customers’ behaviors and
mobile devices [4] and the Internet in every time and every recommend the items or products that they will find the
most relevant among the possible results. Also, they pro-
The associate editor coordinating the review of this manuscript and vide the facilities to enhance the adaption of applications to
approving it for publication was Fabrizio Messina . each user [14]. Recommender systems have been utilized

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
115694 VOLUME 8, 2020
P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

in many fields, like e-commerce [15], health [16], social e-commerce product offering agents issued from 2007 to
networks [17], [18], industry [19], e-learning [20], music 2012. They selected 34 papers to review such highlighted
[21], Internet of Things (IoT) [22], [23], food and nutri- topics as recommendation agent type, preference-elicitation,
tional information system [24], and marketing [25]. They explanation, and the social aspects of recommendation
produce automation of personalization in the e-commerce agents. Also, they covered the operative features of social
environment by employing traditional and modern techniques recommendation agents, variables for user perception such
[26] like machine learning techniques [27]. The recom- as pleasure, comprehended recommendation quality, compre-
mender systems in e-commerce increase the turnover by hended trade-off issue, and comprehended social presence,
providing the relevant products for each customer, sug- and modifying factors, including gender, regulatory focus,
gesting additional goods and products to boost cross-sell reactance level, temporal distance, and decision context. They
and improving client loyalty [28]. The considerable impor- presented the refreshed conceptual model of the recommen-
tance of Internet marketing activities, especially e-commerce dation agent to conduct further research. However, the study
resulting in the personalized recommendation techniques, mostly focused on some issues, such as social presence,
is only one of the features that RSs provide to e-commerce perceived usefulness, trust, satisfaction, and perceived ease
platforms. The main problem is the elevation of the accu- of use as the user’s evaluation factors of recommendation
racy of the RSs results. The vast amount of information agents. Using the conceptual model is a significant point in
that is generated in e-commerce websites makes numer- their research.
ous difficulties for their users to decide which products Lu et al. [34] reviewed the papers published between
they would buy. Employing RSs can greatly address the 2013 and 2015 in the application developments of RSs in
problem [29], [30]. eight basic rankings, including e-government, e-business,
The paper has intended to review recently published papers e-commerce/e-shopping, e-library, e-learning, e-tourism,
in RSs that concentrate on e-commerce and show the devel- e-resource services, and e-group activities [35]. They exam-
oping opinions of these systems, particularly their types, ined RS’s techniques systematically in some perspectives,
methods, algorithms, and implementations in detail. We stud- such as recommendation methods, RS software, real-world
ied some fundamental algorithms, like Collaborative Filter- application domains, and application platforms. Also, they
ing (CF) and knowledge-based, demographic-based, hybrid, used multiple databases for searching papers, such as Science
and content-based filtering. Therefore, we gathered and ana- Direct, ACM Digital Library, IEEE Xplore, and Springer-
lyzed various recent papers of RSs used in e-commerce via a Link. Their review is significantly valuable in both describing
structured methodology [31], [32]. The fundamental goals of the content and categorizing the methods.
this survey are as follows: Furthermore, Karimova [36] reviewed the papers on
• Reviewing the pros and cons of traditional techniques e-commerce recommender system techniques published
that are used in the selected papers. between 2014 and 2016 in journals and conference pro-
• Expressing some of the main challenges of the selected ceedings. He examined the developments of e-commerce
papers and viewing the relevant solutions. recommender systems from the viewpoint of e-commerce
• Pointing out some aspects of RSs to improve their accu- providers or e-vendors. He searched the journal databases
racy and functionality for future studies. like ACM Digital Library, IEEE Xplore, Science Direct,
and SpringerLink. The paper included such critical results
The remainder of the present article is as below: as the dominant role of traditional RS techniques in
Section 2 expounds on the relevant work. In Section 3, e-commerce, especially collaborative filtering and hybrid
we briefly describe RSs in e-commerce, its characteristics, methods, improving the personalized recommendation with
and the relevant metrics. Section 4 discusses the relevant high accuracy, and the researchers’ efforts to overcome prob-
studies in RSs used in the e-commerce field in a Systematic lems, for example, decreasing computational complexity and
Literature Review (SLR) approach. Section 5 expresses the improving recommendation accuracy. The author employed
traditional methods of RSs in the selected papers. Also, the SLR strategy for the review process.
Sections 6 and 7 include the evaluation plan and open issues Li and Karahanna [37] reviewed 40 empirical inves-
based on the review, respectively. In the end, Section 8 con- tigations of the recommendation system published from
cludes the article. 1990 through 2013. Indeed, they classified the selected papers
in three major fields, including understanding consumers,
II. RELATED WORK delivering recommendations, and the impacts of recommen-
Many kinds of research on RSs in e-commerce fields have dation systems. Furthermore, some significant terms like
been performed. The present section explains review arti- customization, interactive decision aid system, personaliza-
cles, which discuss the utilization of RS in e-commerce tion, recommendation system, and recommendation agent
applications and summarizes their main advantages and proposed in the review. The study included some tables with
disadvantages. valuable summarized information in addition to reviewing the
Xiao and Benbasat [33] studied e-commerce product rec- selected papers, and also it refers to a set of theories that
ommendation agents and considered empirical papers on RSs used.

VOLUME 8, 2020 115695

 P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

Adomavicius and Tuzhilin [38] reviewed three main rec-

ommendation technique categories, containing content-based
filtering, hybrid, and collaborative filtering. They provided
constraints and limitations of these methods and possible
solutions to improve the recommendation performance [39].
Sezgin and Özkan [40] selected eight papers in the Health
Recommender System (HRS) domain and provided an SLR
study. They have concluded that HRS is a promising devel-
opment for healthcare services. Also, their review provides a
set of challenges and opportunities in HRS. Khan et al. [41]
have conducted an SLR of 94 shortlisted studies on the field
of Cross-domain recommender systems (CDRS). They have
expressed conventional recommender system problems and
issues in four categories in their selected studies as follows:
• Mature: Accuracy and Sparsity.
• Current research focus: UI modeling, Confidence, Cold-
start, and Diversity.
• Gaining popularity: Trust and Utility.
• Open research issues: Privacy, Novelty, Serendipity,
Risk, and Adaptivity.
FIGURE 1. The benefits of online business based on an e-commerce
Çano and Morisio [42] have presented an SLR on the state of platform.
the art in hybrid recommender systems. Based on their review,
a high amount of selected investigations has combined col- (B2B) [43]. The basic e-commerce advantages result because
laborative filtering with a different method frequently in a it omits the geographical distance and time restrictions.
weighted manner. Additionally, they have concluded that Figure 1 shows some benefits of doing business online by
top problems were cold-start and data sparsity. Also, movie using e-commerce websites, including Personalized Recom-
datasets were widely used by researches to do the hybrid mendation Systems (PRSs) advantages, loyalty, saving the
approaches RS. earth, faster delivery, increase income, easy setup, time-
However, this study aims to review and analyze the selected saving, cost-effective, cost-effective, convenience shopping,
papers based on the following issues: and flexibility.
• Proposing a systematic review and a summary of five
traditional mechanisms in recommender systems in e- B. RECOMMENDATION SYSTEMS CONCEPTS
commerce for emphasizing on the pros and cons in each RELATED TO E-COMMERCE
ranking for the selected paper. The recommender systems help handle user interaction and
• Introducing the challenges of recommender systems sales by assisting users to find products and services that they
mechanisms in e-commerce. may never find by themselves. In traditional commerce and
• Pointing significant issues for future study. business, people have been notified and encouraged about
purchasing products and services through their friends, news,
III. BACKGROUND and marketers [44]. By growing the use of e-commerce and
The present section introduces the main notions and associ- the advancement of its related technologies, the use of RS
ated terminologies regarding e-commerce and recommender has great importance due to its benefits. The recommender
systems. systems usually provide items like products, movies, events,
warnings, and articles to the users, such as customers, visi-
A. E-COMMERCE CONCEPTS AND BENEFITS tors, system administrators, and content. Thus, the approach
E-commerce supports commerce transactions in the most increases the conversion rate and improves the customer’s
uncomplicated approach on the Internet. Although people loyalty and makes satisfaction. Also, based on practical and
consider e-commerce to be online selling and purchasing, reported evidence, RSs can help businesses to improve their
the e-commerce world includes other activity sorts. Thus, vital e-commerce metrics [45]. For example, in a real shop,
any business activity form that is transferred electronically is a good salesman knows the personal preferences of buyers,
called e-commerce. Some instances for e-commerce are elec- and her/his high-quality recommendations make clients’ sat-
tronic payments, online shopping, Internet banking, online isfaction to increase earnings. Some researchers utilize social
ticketing, and online auctions. E-commerce may be cat- resources to get more accuracy in resulted recommendations.
egorized depending on the member type in the transac- For example, Li et al. [46] have incorporated the recommen-
tion in some categories like Business to Consumer (B2C), dation trust, social relation analyses, and preference similar-
Consumer to Consumer (C2C), and Business to Business ity to provide product offerings in e-commerce. They have

115696 VOLUME 8, 2020

P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

shown that their proposed method could outperform better to increase sales and improve customer loyalty. Since the
performance and provide more accuracy than other traditional measurement accuracy is complicated, the authors consider
collaborative filtering based on experimental results. the content of the selected papers.
Also, some studies provided methods to preserves cus-
tomers’ privacy. For instance, Rosaci and Sarné [47] have pre- 2) DIVERSITY/NOVELTY/SERENDIPITY
sented a fully decentralized trust-based B2C recommender RSs use a diversity of items in the recommended list to
system framework for generating more useful recommenda- improve user satisfaction. In other words, when there is a
tions. Their approach utilizes recommender methods of CF higher versatility of intra-list, users have the willingness to
and CB for supporting the traders of the B2C by evaluat- attain more satisfaction with recommendations [60]. There-
ing the reputation of the products, clients’ orientations, and fore, diversity is essential for improving customer loyalty.
the trustworthiness of traders. They have provided a dis- Serendipity is a measure that the RS has been able to surprise
tributed architecture to increase the computation efficiency the user by producing and presenting serendipitous items
of trust measures and recommendations. Thus, the method [61]. Thus, for example, in e-commerce, the system can also
guarantees scalability and privacy. Also, Jiang et al. [48] include one or more attractive items for the user in his/her
have suggested a slope one algorithm depending on the inte- suggestion list. In this case, the recommendations will not be
gration of user similarity and trusted data to perform more bored.
precisely compared to other customary slope-one methods On the other hand, the common investigation is that a
using the Amazon dataset. Selecting trusted data, calculating system might desire to attempt to assess the possibilities,
the similarity between users, and adding the resemblance which a user will identify an item [62]. Novelty is the dimen-
to the weight factor of the developed slope one algorithm sion for measuring the recommendation’s non-obviousness.
are three critical phases of the method to produce the A serendipitous recommendation assists the user in detect-
recommendations. ing a surprisingly exciting item that she/he may not
The recommender system makes better results when it discover.
knows more about the properties of users and items [49].
Also, the system gathers more data from available resources 3) INDEPENDENCE
to feed the algorithms and produces more relevant content for Since the operation of the software is embedded in the current
users. Various evaluation metrics are applied to the systems software application, the approach of using the agent can
to improve the quality of recommendations. The evaluation of be valuable in many applications. Therefore, the metric is
recommendations on impact, interaction, and conversion for called independence, which specifies the proposed method
a particular period are well-known parameters for examining as a part of the current e-commerce software, or it can work
the effectiveness of system output, especially in e-commerce independently or as the agent.
platforms [34]. There are some metrics for evaluating RSs,
like MAE (Mean Absolute Error) [50], MSE (Mean Squared
4) OPERATION COST
Error) [51], RMSE (Root Mean Squared Error) [52], eval-
uating lists of recommendation (based on relevancy levels) Operating costs are expenses associated with running, main-
[53], recall [54], MAP (Mean Average Precision) [55], nDCG tenance, and administration of a recommender system in
(normalized Discounted Cumulative Gain) [56], diversity this paper based on computational resources such as CPU
(intra-list similarity) [57], implicit feedback, diversity of consumption, memory usage, and network traffic. In other
Lathia et al. [58], and MPR (Mean Percentage Ranking). words, this metric indicates whether the method needs to
be optimized or not. Some authors emphasized that their
C. METRICS DEFINITIONS methodologies must be optimized for real word usages.
The current study illustrates an RSs technique comparison
with the utilization of such measures of performance like 5) RESPONSE TIME
security, response time, scalability, accuracy, operation cost, When the online user demands a recommendation, she/he
diversity, novelty, serendipity [59], type of data sources anticipates to get the response abruptly, therefore, measuring
(implicit/explicit), and evaluation method. Also, we com- the response time that the user tolerates is very important to
prised and evaluated the selected papers in section 6 by con- evaluate RSs [63].
sidering these metrics. Consequently, this section contributes
to a brief discussion of them. 6) SCALABILITY
A scalable system is a system that is capable of meeting
1) ACCURACY the growing demands of work or process and is capable of
Accuracy is known as a part of the true recommenda- integrating growth [64]. For example, when we call a recom-
tions from the whole feasible recommendations. In other mender system technique scalable, it can be applied to large-
words, the better quality of the output recommendations scale input data without compromising on the efficiency of
leads to having a higher accuracy of the system and fewer the technique or forcing any considerable overhead on the
errors. In e-commerce, the quality of recommendations led system.

VOLUME 8, 2020 115697

 P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

7) SECURITY • RQ 2: What are the open issues?

RSs collect more and more such personal information as login Section 6 explains these matters.
time, location, and social network profiles and data [65] to • RQ 3: How are the search procedure of paper and the
provide better recommendation services, so they make high selection of it for assessing?
privacy concerns to the public. On the other hand, RSs have The answer is provided in Section 4.
to protect themselves from attacks begun by malicious users • RQ 4: How researchers conducted the study?
who want to influence the recommendations [66]. So, it is In section 6, we directed the relevant answer.
essential to consider the security issues in the RSs. • RQ 5: What parameters are necessary for this research,
and what are their assessments from the viewpoint of the
8) TYPE OF THE DATA SOURCES (IMPLICIT/EXPLICIT) researchers in this study?
The input data of the algorithms used in the recommender The second part includes a response to it.
systems affect the quality of the resulted recommendations.
Some of these algorithms require the use of explicit data, B. PAPER SELECTION PROCESS
such as user rating matrix, users’ profiles, and items’ cat- To identify the relevant papers for completing the review plan,
alogs. However, others work with implicit data only like in the first step, we used the Google Scholar service to do the
user behavior. Moreover, some methods use both types. The searching phase. Thus, we searched the papers for various
use of implicit data requires some particular processes and combinations of the relevant keywords, including recom-
methods, such as the use of data mining and machine learning mender system(s), recommendation system(s), e-commerce,
techniques [67]. Therefore, in this research, this metric is used and electronic commerce. We also applied the ‘‘allintitle’’
to classify these types of algorithms. filter of the Google Scholar service to find the most relevant
studies.
IV. RESEARCH METHOD We did our search on April 20, 2019, and considered the
To present a clear image of the techniques in recommender range of publishing from 2008 to 2019. It is important to
systems in e-commerce, we provided a Systematic Literature note that the titles of some research directly point to the
Review (SLR) along with a concentration on the relevant recommender systems in the e-commerce environment, while
and recent studies. Thus, we used the applied review pro- some titles are related implicitly. Therefore, we considered
tocol based on the guidelines of Kitchenham and Charters research whose title was directly related to the subject of the
[68] and considered the implementation of an exhaustive study. Table 1 shows the number of papers found according
and thorough abstract of the present literature related to to the search terms. The total number of detected items was
the field of the study as one of the SLR purposes [32], 289 papers by eliminating the patents and citations types.
[69], [70]. For reducing the researcher’s bias, the use of
review protocol is inevitable. As such, we focused on a set TABLE 1. Results from google scholar service related to search keys using
of Research Questions (RQs) that serve as the object of the ‘‘allintitle’’ phrase.

this literature and originate from the ideas that initiated this
review. We completed this study by conducting three pro-
cesses, including question formalization, searching papers,
and applying the desired limitations to select the relevant
papers to review. Consequently, we outlined the research
aims as RQs in Section 4.1, searching papers process from
resource libraries in section 4.2, and finally, the classification
and inclusion/exclusion criteria are applied in Section 4.3.

A. QUESTION FORMALIZATION
The objective of the existing research is to collect all of the
credible and influential studies that investigated the recom-
mender systems in e-commerce in recent years. In particular,
exploring the techniques, open issues, and critical points of
the selected papers are the main aims of this paper. To achieve After the searching phase, we applied our selection strategy
these goals, we defined the Research Questions (RQs) as based on some crucial issues, such as the year of publica-
follows: tion, the language of the papers, and their title. We limited
• RQ 1: What are the recommender systems in e- our study to English papers. Also, recently published papers
commerce and their important and applied techniques? were of greater importance for selection, and we consider
Which types of research methods are useful? What cat- the reputation and validity of the journals. Figure 2 shows
egories are important? the distribution of the papers obtained by publication year
Section 5 conducts the response to the mentioned from 2008 until 2019. In 2014, the published papers had the
question. highest rank.

115698 VOLUME 8, 2020

P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

FIGURE 2. The number of papers published from 2008 to 2019 in the

subject of the study area.

FIGURE 4. A pie chart of classification of selected papers among

publishers.

FIGURE 3. A pie chart of the percentage of the papers based on selected

publishers (Step 1).

Then, we excluded some types of studies, such as the-

sis, reports, work articles, commentaries, editorial notes,
books, and books’ sections. Additionally, we did not con- FIGURE 5. Distributions of selected papers over time infamous
publications for reviewing.
sider non-English papers. To select the most relevant papers,
we reviewed the chosen papers considering the year of the
publication, subject, and journal category as the primary crit- Figure 5 shows the distributions of selected papers over
ical points for the final selection step. For example, we con- time infamous publications to review. In the 2016 portion,
sidered some issues related to journal filtering, such as having it shows the highest rank.
high-quality factors in journal publications, the validity of
retrieved items, novelty, and most relevant results as well as V. REVIEW OF SELECTED PAPERS
the critical points found in the conclusion sections. In addition to solving the information overload problem,
Figure 3 illustrates the paper ranking by various publishers. the RSs help users to find products related to their preferences
We considered all the studies which are published by famous by providing personalized services [38]. We want to review
publishers and removed the rest from our field of study. After the basic models of recommendation techniques to examine
purifying the items, we selected 33 papers to review and recommender systems in e-commerce. The user-item interac-
examine them. tions, like ratings or purchasing behavior, and the information
Finally, the categorization of the articles chosen depending for quality regarding the items and users like related keywords
on the basic recommender system methods in e-commerce or textual profiles are two fundamental kinds of input data to
is described. Table 2 demonstrates the categorization of feed basic models of RSs. CF methods use the user-item inter-
the related investigations in all the groups. 11 papers out actions [101], and CBF approaches employ the features of
of 33 papers are based on the CF method. Also, six papers users and items [102]. Explicitly specify the outward knowl-
are related to DBF, three papers are in the CBF category, edge foundations, user requirements, and restrictions that are
and for KBF and hybrid, we selected five and eight papers, essential information for KBF methods [103]. Demographic
respectively. RSs use demographic information regarding the user for
Additionally, Figure 4 illustrates the paper classification constructing the recommendations based on special mapping
among seven publishers. IEEE publishes 54% of the arti- demographics for ratings or purchasing tendencies [104].
cles, and Springer publishes 17% are on the top of other Hybrid systems are able to integrate the power of different
publications. RSs kinds for creating systems that may operate powerfully

VOLUME 8, 2020 115699

 P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

TABLE 2. Selected papers distinguished by categories and publishers to review.

in an extended diversity of settings [105]. Also, Bajpai and A. COLLABORATIVE FILTERING MECHANISM
Yadav [106] provided an improved dynamic e-commerce In this section, first, CF mechanisms and their fundamental
RS by using web usage mining and content mining. They aspects have been described in Section 5.1.1. Afterward,
examined the proposed method and concluded the accuracy the selected papers are reviewed in Section 5.1.2. Ultimately,
of the system in terms of precision, recall, and memory in Section 5.1.3 provides and compares their problems,
consumption in comparison with other traditional models. advantages, and disadvantages.
Also, based on the definition of the selected metrics that are
expressed in section 3.3, we evaluated the selected papers. 1) OVERVIEW OF COLLABORATIVE FILTERING MECHANISM
In this section, we examine the selected papers in basic CF, one of the most popular and widely implemented methods
categories of RSs. Also, we review their methods, differences, in RSs, is a technique utilized for linking the data of an applier
benefits, and challenges. with the data of other alike appliers depending on buying

115700 VOLUME 8, 2020

P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

patterns to produce directions for the user for prospective indicated that the model-based RS performs better com-
shopping [71], [75], [85], [107]–[114]. Amazon uses CF pared to memory-based RS in the precision, the calculation
techniques for making its recommendations depending on time, and of recommendation relevance. They concluded
a client’s previous buying and buying of those that bought that model-based CF is better compared to memory-based
the same products. E-commerce platforms use this technique ones concerning calculational velocity in products of rec-
to create market segmentation depending on client behavior ommending. Also, they resulted in the model-based CF is
compared to the psychographics and demographics metrics. better compared to memory-based ones in producing the
Creating a database of user preferences and processing it recommendations depending on the mean number of related
optimally for the preparation of recommendations is the products comprehended by the applier. The study did not con-
primary performance of this method [115]. Scalability and cern with some other features of RS performance, like variety,
accuracy are two different critical issues associated with the serendipity, coverage, and newness. Also, they implemented
CF technique. The problems of scalability and efficiency on model-based and memory-based CF by using developed
a global scale require modern processing space and speed Naïve Bayes and Nearest Neighbor techniques.
optimization to support customer satisfaction. Thus, CF is Wu et al. [71] studied an improved CF-based recommender
the most popular and successful technique that utilizes client system for mobile e-commerce to provide personalized rec-
ratings, details, and reviews gathered from the total clients to ommendations. The method comprises three primary oper-
construct recommendations [116]. Several studies presented ations: recommendation method module, output functions
the proper framework [117]–[122]. module, and input function module. Also, they stated that
There have been two principal modes of implementing some problems like the attainment of data depended on the
collaborative filtering: a) memory-based approach, b) model- clear assessment by the user, insufficiency in spontaneously
based approach [123]. The memory-based approach com- attaining obvious data of the user, and rarity of the infor-
putes the resemblance among the appliers by computing the mation in system progress systems. Thus, they designed the
similarity function like cosine formula. The model-based Item Rating Prediction-based CF method to firstly prognos-
approach uses some sophisticated methods like machine ticating the client rankings for items unrated by the calcu-
learning techniques to find patterns in the dataset and learn lation of the resemblance among items and then determine
from them to employ the new data [124] and also some the closest neighbors by attainment of a novel resemblance
approaches like matrix factorization [125]. measuring procedure. Furthermore, they studied the item
rating prediction-based CF to solve the specific sparsity of
2) REVIEWING SELECTED COLLABORATIVE the user data problem in mobile electronic business. Finally,
FILTERING MECHANISMS they suggested some issues required for ongoing progress and
Sivapalan et al. [116] showed the significant challenges completeness in pragmatic utilization for future studies.
of CF methods such as sparsity, scalability, and synonymy Cao et al. [76] designed and implemented a distributed
and expressed high personalization for users enabled by SVD++ algorithm for enhancing the recommendation qual-
using RSs. The CF method uses the user ratings, reviews, ity of the personalized items in an e-commerce platform. The
and details gathered from the total appliers to produce rec- SVD++ algorithm can use implicit and explicit feedback
ommendations by analyzing the active user with the same data of the client at the same time. Thus, they decided to
characteristics and preferences of the present applier. More- evaluate the distributed calculation of the SVD++ algorithm.
over, the CF approach applies some resemblance measures, Also, they made use of the algorithm implemented in the
containing cosine metric, personalized data, and Jaccard real exam in the products recommended. They concluded that
coefficient. They concluded that due to data sparsity, sales their proposed algorithm solves the problem of mass data
volumes on large e-commerce sites are decreasing. There- processing with the data size up to TB level or scalability
fore, they suggested that the developers of the RSs stop problem, and the difficulty of the sparse matrix to a deter-
using the nearest neighboring algorithms in implementing the mined limit, and notably improve the personalized product
RSs for large e-commerce websites. Because they make low recommendation quality. Although this method has some sig-
accuracy in recommendations, and also the coverage metric nificant advantages, it requires high processing costs. It also
reduces. Also, the similarity measurement of product or users has a particular complexity due to the use of big data technol-
is the essential neighborhood-based CF procedure compo- ogy and distributed data platforms.
nent. Some measures like Pearson’s Correlation, Cosine, and Lin and Wenzheng [74] designed and implemented a per-
Adjusted Cosine often used for the CF method [126]. Some sonalized recommender system in e-commerce depending on
studies expressed new similarity measurement techniques for the web mining approach. As the web data such as content
specific criteria in CF. For example, Patra et al. [127] showed and structure have high complexity, the use of data mining
a new one for sparse data. or other technologies for providing them as the input data
Aditya et al. [75] examined the memory-based and model- for use in e-commerce recommender systems has a certain
based CF performance as two implementation plans using a complexity. The system considered the main functionality of
data sample of an e-commerce case study. They evaluated any e-commerce project, such as user, commodity, and order
each method by using offline and user-based testing and management to provide the appropriate recommendations.

VOLUME 8, 2020 115701

 P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

However, the performance of the proposed method is related prove the performance of it. They showed that the stated sys-
to computing resources such as CPU power. Thus, the opti- tem produces further efficiency compared to the conventional
mization and development of performance are necessary for collaborative filtering system regarding buying and customer.
further studies to apply the method for implementing RSs. Also, the replacing suggestions are always better compared to
Sun et al. [73] proposed a big-data based retail RS based supplementary offerings for buying and customers.
on CF by using the corresponding distributed computing Jiang et al. [79] proposed a slope one algorithm depend-
algorithm on MapReduce. The method used cosine resem- ing on the integration of user resemblance and valid data.
blance and Pearson resemblance correlations in the Item-CF Their method uses three procedures: trusted data selecting,
algorithm. The importance of this research is that customers calculating the similarity between users, and including the
do not rate any items in real stores. Hence, the authors of resemblance to the factor of weight of the elevated slope one
the article examined the use of RSs for marketing activities algorithm. Also, they used the Amazon dataset to show the
such as selecting the most relevant store-product pairs and more accurate performance of the proposed method than the
deploying product advertising. They concluded that, in addi- customary slope one algorithm.
tion to the effectiveness of the system on retail sales regarding Gaikwad et al. [80] proposed a model-based CF recom-
goods and shops, it could deal with large-scale datasets and mender system depending on the eventual model with the
get further scalability and a new mean of accuracy marketing utilization of developed Naive Bayes algorithm. They used
assistance. Because of the requirement to use the concept of query time, search query, and click time as the aspects of the
big-data technology, this method has high computational cost model of the Naive Bayes algorithm. They resulted that their
and complexity. approach produces more precision than a basic Naive Bayes
Choi and Kim [72] proposed a recommendation mecha- model.
nism using the iteration purchase number for goods for each
client as the repurchase ratio measurement against a one- 3) COMPARISON AND SUMMARY OF THE
time purchase as the recommendation criteria and explored MECHANISMS REVIEWED
the repetitive purchase pattern. The repurchase ratio is one In the previous section, we analyzed eight selected papers
of the critical metrics of customer loyalty in business. The in CF mechanisms and their pros and cons. We found that
method offered three product groups by making use of a some issues, such as the cold-start problem, low security, low
user-based CF method, item-based CF procedure, and by trust, and high response time, are the main problems in the
discovering Associate Products, which customers often buy CF methods.
them simultaneously. Thus, they used the user-based CF and Table 3 illustrates the proposed methods and the compar-
item-based CF method to recommend the associate items ison of the most significant pros and cons of each article.
analyzed by Association Rules. Also, they examined the sales The problems with Low Scalability, Data Sparsity problem,
data belonging to a firm for almost one and a half years to Data acquisition problem, and high computational time are
obtain the efficiency of the proposed method. unsolved entirely in this mechanism. The method has advan-
Also, Kuang [77] proposed the recommender systems in tages like better relevance, recommendation quality result,
e-commerce, depending on the CF mechanism. Furthermore, and considering the user experiences. Although efforts made
due to a large number of users in real-world applications by researchers to find solutions for these problems, this
of e-commerce, the researchers should optimize algorithms method still requires further development.
in such a way to minimize the time it takes to respond to
each user request and time spent. Despite the conventional B. DEMOGRAPHIC-BASED FILTERING MECHANISMS
algorithms, the effectiveness and reasonability of the pro- In the current part, first, DBF mechanisms and their fun-
posed method in the accuracy of prediction of interest degree damental aspects are explained in Section 5.2.1. Afterward,
verified via the experimental datasets. He used the prognosti- the selected papers are reviewed in Section 5.2.2. Ultimately,
cation and real amounts of the Mean Absolute Error (MAE) their problems and pros and cons are compared and consid-
as a standard measure for algorithm precision. Also, running ered in Section 5.2.3.
the CF algorithms in e-commerce websites with a significant
amount of data related to the products’ and users’ profiles 1) OVERVIEW OF DEMOGRAPHIC-BASED
have a high amount of the computational resources require- MECHANISMS
ments to process in the RSs algorithms. The DBF algorithms apply the users’ demographic profile
Hwangbo et al. [78] offered a procedure to suggest fashion such as location, language, age, religions, and some useful
goods for clients by enhancing the existing CF procedure to parameters to accomplish the items’ recommenders to the
consider the features of fashion products. They also assessed users. This technique makes the recommendations to be more
offline and online modes of buying those products in their relevant to the user realities [128]. Thus, it is widely used by
approach. Also, they recognized that the product that the e-commerce companies as an effective solution to respond to
client wants to buy is considered as a product that substitutes their user interests. Some RSs are based on the demographics
or completes the product that the consumer favored earlier. of customers [28], [129], and it is necessary for e-commerce
They employed the system to a real online shopping mall to environments.

115702 VOLUME 8, 2020

P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

TABLE 3. Comparing the most significant pros and cons of the CF mechanisms of selected papers in recommender systems in e-commerce.

2) REVIEWING SELECTED DEMOGRAPHIC-BASED personalized recommender system in e-commerce that con-

MECHANISMS sists of four phases: 1) Acquire information explicitly and
Cho and Ryu [81] proposed an applied e-commerce personal- implicitly, 2) Preprocessing of the information, 3) Recom-
ized RS making use of the RFM method and the demographic mendations of the form, and 4) Indicating recommendation
data. RFM analysis involves the use of three key customer outcomes to clients. Plus, they employed a data diversity
behavior metrics, including Recency, Frequency, and Mone- in e-commerce such as server log data, online market data,
tary. Recency is the number of days since the last purchase query data, hyperlinks inside web pages, web pages, and
of the customer. Frequency is measured as the number of client registration data. Finally, the authors concluded that
orders placed within a specified period, and Monetary is the the proposed method improves the quality of recommenda-
total amount of money spent by a customer in a specified tion efficiently. However, although the authors of the arti-
time. On the other hand, they used several data groups ranked cle presented the proposed approach concerning the great
by demographic variables, including gender, age, customer’s theoretical and practical significance and some advantages
propensity, and vocation for the user’s personal information. such as providing real-time recommendations and enhancing
Also, they showed the procedure of the proposed system the recommendation quality, they did not show consideration
and given some technical concerns about its performance about the privacy and the security.
by using the implicit data without causing trouble for the Souali et al. [83] introduced an ethical-based RS using
user. Also, the system contains three agents: 1) analytical the data related to behaviors and customs with no interven-
agent, 2) the recommending agent, and 3) the learning agent. tion by the user. The method used the information written
They implemented for prototyping of the internet shopping by clients within their registration level and depending on
mall that works for the cosmetics professionally [76]. The a catalog of the ethics. They expressed that they want to
authors solved the new user or new item challenges using develop an enhanced RS that automatically supports ethics
the clustering purchase data approach. Finally, the method without the user intervention to examine the efficiency, accu-
has recommended the item with a high probability of pur- racy, and some potential challenges. The significant problem
chasing and has increased the accuracy of recommendations with the use of DBF recommendation methods is that some
experimentally. individuals with similar demographic features may different
Fan et al. [82] used the customer registration informa- interests. The detection of these differences is not possible by
tion or demographic characteristics of users as an essential this method. Therefore, the researchers recommended using
source of data mining algorithms for implementing RSs in the DBF method in the hybrid approach. However, despite
e-commerce websites. They proposed the workflow of a the high level of accuracy offered by the recommendation

VOLUME 8, 2020 115703

 P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

TABLE 4. Comparing the most significant pros and cons of the DBF mechanisms of selected papers in recommender systems in e-commerce.

results, it works poorly for new users because of the lack of operation costs and highly complex systems are two signifi-
demographic information. cant disadvantages. The author suggested using a heuristic
search and artificial intelligence (expert system) implemen-
3) COMPARISON AND SUMMARY OF THE tation method for future research for this domain.
MECHANISMS REVIEWED Chun and Hong [130] presented a KBF product recom-
In the prior part, we analyzed six selected papers in DBF mendation and implemented the proposed system using JESS
mechanisms and their pros and cons. The selected papers (Java Expert System Shell) and Java servlet and examine it
mentioned that the main problem in the DBF method is on a sample internet shopping mall that sells cellular phones.
data collecting about demographics. We need to do some Also, the architecture of the proposed system consists of four
preprocessing algorithms to achieve the related data. sections: an inference engine, a user interface module, a client
Table 4 shows the proposed methods and the comparison database, and a knowledge base of the product field. Given
conducted between the most significant pros and cons of the that the system was designed for the phone store, it stores and
articles. One of the optimal choices is to use this method to processes related data for this domain, namely score, picture,
solve problems such as over-specialization problem and new model, Service Company, manufacture, and price, so that
user problem. This technique has unique flexibility for use in it can generate related proposals. Information gathering fed
different domains and increases cross-sell. With all of these on questioning the user and the system find the products
benefits, issues such as the lack of attention to significant that meet the user’s requirements best. The product domain,
ethical challenges, the problem of user new habit and security, inference engine, and user interface module are required
are still not entirely solved in this approach. parts of knowledge based on the system. Since the use of
hybrid methods compared to the single plan is a better result,
C. KNOWLEDGE-BASED MECHANISMS the authors of the paper suggested the implementation of the
In the present part, first, the mechanisms of the KBF in RSs hybrid method, including CF and KBF, using JESS in the
and their fundamental aspects are explained in Section 5.3.1. future studies.
Afterward, the selected papers are reviewed in Section 5.3.2. Knowledge-based methods require the creation of a
Ultimately, their problems and pros and cons are compared database containing information related to the goods or items
and explained in Section 5.3.3. in the area of use. For example, in a car shop, features like
type, size, and color are stored in the database, and offers
1) OVERVIEW OF KNOWLEDGE-BASED MECHANISMS based on these features are provided to users. On the other
KBF recommender approach is depended on explicit hand, the combination of the method with CF usually pro-
knowledge about the criteria of the recommendation, item vides better recommendations. Therefore, Tran [131] intro-
classification, and user preferences. In KBF RSs, a mea- duced a hybrid RS architecture that uses the integrating of
sure of distance or similarity utility like Euclidean distance KBF and CF techniques. Although using the KBF method
and cosine similarity judges how much the recommendation does not depend on the user ratings of items, it is essential
match with user needs. Although solving the cold start prob- to know the features of the relevant field, so the weaknesses
lem is one of the strengths of this approach, the difficulty of of the approach are the necessity of knowledge engineering,
understanding the need to define recommendation knowledge which is a difficulty.
explicitly is its weaknesses. Also, to solve the fake neighborhoods in e-commerce
RSs, Martín-Vicente et al. [85] suggested a method depend-
2) REVIEW OF SELECTED KNOWLEDGE-BASED ing on applying semantic logic for the CF mechanism.
MECHANISMS They decreased the size of the user interests’ vector greatly
Ya [84] proposed using the expert system technology for to reduce the computational cost of the process. Also,
implementing the personalized recommender system in the authors showed the improvement of the quality of rec-
e-commerce. The proposed method solved the complex ommendations. Two primary disadvantages of the proposed
e-commerce personalized recommendation problem, but high method are high data sparsity problem and low scalability.

115704 VOLUME 8, 2020

P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

Using the lab experiment, Ku et al. [86] outlined the sig- tested respectively on mainly model of the front and back
nificant attributes of e-commerce product RS as the important systems using a series of test data, and the results of experts
hints for designing RS. They chose and examined six signif- recommended the decision algorithm model to validate the
icant attributes of RS concerning users’ attitudes and prefer- feasibility of the proposed model. Additionally, the method
ences toward an RS interface. These features are description, has examined in four levels: data collection, user modeling,
rating, associated products, offered product number, client’s the algorithm of recommendation, and recommend output.
review, and a comment of the review. They found that the The proposed plan keeps the user model up to date in RSs
feature ‘‘Explanation for Recommendation’’ is significant by collecting and analyzing the user information regularly
for all participants. ‘‘Rating’’ is essential for the particular to build the user model and then rebuild the model after
members, and ‘‘Comment’’ looks to contain less important examining the feedback. The proposed model overcomes the
for all of them. They adopted a conjoint analysis for examin- new- user challenges because it will suggest the best-selling
ing the differences in the clients’ preferences for the quality merchandise for new users. Also, for registered users, it uses
integration of the interface of the RS. Also, they measured the user characteristics resulted from some user-related data
the significance ranking of RS’s properties in various parts like age, occupation, interests, and hobbies.
when RS appliers have been looking around the offered con-
tents. Despite the valuable advantages of the method, like 3) COMPARISON AND SUMMARY OF THE
enhancing the satisfaction of the client and producing user MECHANISMS REVIEWED
trust, it does not solve the scalability problem. The RAs track In the prior part, we analyzed six selected papers in the
online user action or collect data from the customer explicitly mechanisms of the KBF and their pros and cons.
to provide high-quality recommendations about the services Table 5 illustrates the proposed methods and the compari-
that the client may like them. son of the most significant pros and cons of the articles. Some
Huseynov et al. [89] assessed the influence of KBF problems like a cold-start problem, early-rater problem, and
Recommender Agents (RAs) on the online-client decision- fake neighborhoods are solved using this method. Although
making procedure, where the RAs track. they considered the they bring customer satisfaction and, in some cases, boost
length of the shopping, buying the necessary item, and the sales and enhance the recommendation quality, the problems
attempt to look for the essential products. They showed that such as low scalability and high operation cost still show up.
KBF RAs enhance the client decision-making method by
decreasing the length of the shopping and power used in look- D. CONTENT-BASED FILTERING MECHANISMS
ing for proper products. Furthermore, they concluded that the In this section, first, the CBF mechanisms in RSs and their
technique improves the quality of the decision, and increases fundamental properties are explained in Section 5.4.1. After-
the customer number who buy the items they want. Although ward, the chosen papers have been reviewed in Section 5.4.2.
the proposed method facilitates the decision-making process In the end, in Section 5.4.3, their problems and pros and cons
for users to purchase, authors of the article tested and evalu- have been considered and compared.
ated it in a limited environment by university students, and to
ensure results, and the system must be in a real situation with 1) OVERVIEW OF CONTENT-BASED FILTERING
regular e-commerce users. MECHANISMS
Luo [87] proposed a framework of Web mining based on The RSs use the customer preferences and description of
the data clustering analysis, including content, web usage, the items to filter the query resulting data and suggest the
and structure, to provide a high-quality recommendation for proper items to the user. Fundamentally, the CBF techniques
recommender systems in e-commerce. Moreover, He used recommend items or products which are similar to the items
pattern analysis in web mining to store knowledge rules on that user liked priorly or is looking for nowadays [84], [90].
proper media and support the recommender engine. Finally, In this case, items have been grouped by how closely they
the author concluded that the proposed method could deal match the user attribute profile, and the best ones are offered.
with large data volume, improving scalability, and response On the other hand, the CBF algorithms recommend the items
speed considerably. On the other hand, the system integrates that are like those the user liked priorly, but the CF methods
all the results from web usage, content, and structure mining. recommend the items that similar users liked.
It provides the quality personalized RSs in the cases where
there is fewer usage data or frequent change of website 2) REVIEW OF SELECTED CONTENT-BASED
content. With significant advantages such as improving the FILTERING MECHANISMS
accuracy of resulted recommendations, the system still has Gatchalee et al. [90] described an ontology development for
some disadvantages, such as complexity and high computing small and Medium-Sized Businesses (SMEs) e-commerce
cost. Plus, collecting various web data, such as structural, website depended on the analysis method of the con-
content, and usage data are challenging, complicated, and tent for acquiring knowledge for developing RS based on
expensive. this ontology. The proposed system consists of three main
Xiaosen [88] has analyzed and designed the requirements tasks: 1) content analysis, 2) Ontology development, and
and function of personalized RS for e-commerce. He has 3) Recommendation system for SME e-commerce websites.

VOLUME 8, 2020 115705

 P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

TABLE 5. Comparing the most significant pros and cons of the KBF mechanisms of selected papers in recommender systems in e-commerce.

The authors suggested that content analysis plays the pri- another area like movie or music suggestion concerning a
mary role in developing ontologies of any domain for distinct life scene and novel time decay factor for future work.
improving RSs. They defined the ontology domain as the Palopoli et al. [93] proposed a distributed architecture
SME e-commerce website in Thailand with the significant of RS depending on a system of the multi-tiered agent to
concepts of the ontology. The proposed method recom- generate more useful recommendations and reduce the oper-
mends sample websites with considerable relevance to meet ation cost. This plan overcomes the calculational difficulty
user requirement rules. However, since SME e-commerce on the gadget that the local agents operate, and the exis-
sites operate independently and different structures, col- tence of determined agents enhances the clients’ knowledge
lecting their essential information and analyzing them with depiction. They conducted practical operations to assess the
changes over time may be problematic to provide the online system performance concerning benefit by executing spe-
recommendation. cific software. Also, the method has significant advantages
The CBF techniques use structured and unstructured text concerning openness, privacy, and security. The authors sug-
documents, multimedia such as images, photos, songs, and gested the principal issues, such as considering the various
other file formats to extract features. Recently, some studies moral models of a B2C procedure and producing behaviors,
introduced a sentiment analysis of contents and different examining other related features in a distributed environment
machine learning algorithms to gather some useful inputs like dependability, protection, and reliability.
for recommendation components. Gao et al. [91] expressed Miao [94] has proposed the effective mobile recommender
that the limited resource is the primary challenge of the systems in e-commerce method. The method combines CBF
CBF approach and challenges of new users as well as other technology with CF technology to get high accuracy in
methods. Moreover, limited information or tags reduces the recommendations. Additionally, the author has introduced
quality of recommendations. a weighted combination filtering RS that mainly consists
Because of problems such as sparse and missing in the of three modules: the input function module to collect
use of explicit information, the researchers try to use implicit required information about customer’s personal and com-
information to explore user activities such as purchase, click, munity groups, the core function module of RS to generate
collection, and cart. Chen et al. [92] combined some types recommendations, and the output function module to provide
of user actions and timestamps to get better recommendation the content for users. Meanwhile, the latest module uses the
results. They introduced the At-BPR method to overcome the results of two sub-sections: the suggestion component and the
restriction of one kind of action in Bayesian Personalized prediction component.
Ranking (BPR) and other extensions to explore user pref-
erences and got better-ranking performances experimentally. 3) COMPARISON AND SUMMARY OF THE
They used a global score function on the proposed method to MECHANISMS REVIEWED
assess the preferences of the user over various items. On the In the prior part, we analyzed six chosen papers in CBF
other hand, the authors employed two different sampling mechanisms and their pros and cons. Table 6 illustrates the
strategies in the experiments as an alternative to consis- comparing procedure of the significant pros and cons of the
tent sampling like BPR adoption. They suggested exploring selected mechanisms. As the table shows, the author could

115706 VOLUME 8, 2020

P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

TABLE 6. Comparing the most significant pros and cons of the CBF mechanisms of selected papers in recommender systems in e-commerce.

consider some more complex parameters to get better results technique using MapReduce to speed up the execution pro-
in their methods. Also, this point needs to be worked on cess of RS algorithms and cloud technology, highlight the
the optimization of the proposed algorithms. Although this system performance, and solve the scalability challenge. The
method solves the problems like cold-start and the complex authors employed several RS algorithms to use in various
e-commerce personalized recommendation, it improves the web pages of real e-commerce applications to meet various
quality of resulted recommendations compared to other meth- necessities of the user. They tested the method on a real
ods. However, it still has problems such as increased process- platform and showed an increase in system performance.
ing resources. On the other hand, to enhance the accuracy of They suggested the improvement of parallel processing opti-
this method, we need to consider more parameters, which mization for future studies.
increases the response time and the need for storage and The hybrid approach in RSs is the use of several differ-
processing resources. ent algorithms for collecting and processing information of
the users or the items that improve the quality of recom-
E. HYBRID MECHANISMS mendations. Zakharov and Philippov [99] proposed a hybrid
In the current part, first, the hybrid mechanisms in RSs method for data preparation for recommender systems in
and their fundamental aspects have been explained in e-commerce. They solved the cold-start problem using the
Section 5.5.1. Afterward, the selected papers are reviewed in Item-Item CF algorithm and User-User CF. Explicit and
Section 5.5.2. Ultimately, their problems and pros and cons implicit user data clustering is the base of both algorithms.
have been compared and considered in Section 5.5.3. They concluded that their proposed method could raise the
number of resulting products in fewer clicks than the plans
that do not make utilization of the implicit data. The system
1) OVERVIEW OF HYBRID MECHANISMS
produces a simple environment and empowers the visitors’
We use hybrid RS to create a more robust RS by join- fidelity toward the online resources that, in the medium term,
ing various algorithms and methods. Hybrid RS provides may result in increasing the returning observations and the
all the benefits of algorithms and reduces their disadvan- production of the pool of loyal clients, who certify constant
tages effects. For example, by linking CF methods, where yield enhancement of the online shop.
it breaks in un-rating new items, with content-based mod- Also, Xue et al. [97] advanced hybrid RS using the Bipar-
els, anyplace that aspect info on the items is available, tite Network Recommendation model depending on the allot-
novel items may be offered more accurately and more ment of the resource and developed the CF model to optimize
efficiently [39], [96]. the accuracy and coverage of the recommendation results.
They employed a graphing approach to abstract items and
2) REVIEWING CHOSEN HYBRID MECHANISMS users as its nodes and mined purchase information and user
Dong et al. [96] proposed a personalized hybrid RS to click behavior to get user preference ratings. Moreover, the
address the challenges that traditional algorithms face in conclusion has shown the improvement of recommendation
the growth of the number of products and users of the accuracy and customer satisfaction. Furthermore, because
e-commerce platforms, such as improving accuracy in user’s of the shortcomings mentioned in the research, such as the
interests modeling, providing high diversity, and supporting limitation of the experimental use of e-commerce transaction
large-scale expansion. Also, the method can support mas- data and lacking demography and other commodities fea-
sive dataset. However, they designed and implemented the tures, they suggested some open issues for future researches.

VOLUME 8, 2020 115707

 P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

These issues are considering the feeling mining to assess the complete their approach. Also, they compared the well-
client’s preferences, and examining the environmental factors known e-commerce websites like Amazon, eBay, FlipKart,
to conduct the more proper personalized suggestion. Snapdeal, and Paytm in the article. In the approach, they
Qiu et al. [95] studied the complex e-commerce architec- stored better modeling of the item and client information
ture, which can collect multiple data types, use a difference in the database and considered some abundant information
of recommendation methods for meeting the requirements like feedback, and multi-criteria ratings in RSs. The authors
of various recommendation service sorts. They schemed a proved that the proposed method could successfully integrate
multi-mode recommender system in e-commerce to improve recommendations and semantics to improve products or ser-
accuracy and comprehensively of recommendations. On the vices recommendation results.
other hand, the approach collects a wide range of user’s infor- Jiao et al. [98] presented the fuzzy theory to cope with the
mation, combines with a variety of recommendation methods clients’ data of the behavior and classify the result of quanti-
that can learn from each other, and finally comprehensively tative semantic analysis of the trade comment text data. They
consolidates all results into an overall conclusion to provide built the personalized recommendation model and improved
the users’ goods of service recommendations. Moreover, the its accuracy using the applied evaluation and optimization
proposed system solved data sparsity and cold-start problems function with the case study. As both users’ data and items’
and has advanced high efficiency with low response time. data have similar data analyzing and processing, they rec-
Ultimately, although this method solves the above problems, ommended the optimization of the situation. On the other
it has a high operating cost, and some optimizations are hand, they defined their proposed model in five levels, includ-
needed. ing purchase database, customer shopping website, data for-
Aprilianti, et al. [39] proposed and implemented a mat, customer shopping personalized recommendation, and
weighted parallel hybrid method for recommender systems the customer level. The authors showed the growth of user
in e-commerce in Indonesia, which is a combination of the satisfaction using the radar chart and considered the user’s
CF and CBF approaches. Additionally, the proposed plan can experiences by providing high accuracy.
overcome the weaknesses of those two methods, such as the
item diversity shortage to be offered to the client in CBF, and 3) COMPARISON AND SUMMARY OF THE
the cold-start issue in CF [132], [133]. However, the disad- MECHANISMS REVIEWED
vantages of the system are the impossibility of providing the In the prior part, we made an analysis of eight chosen papers
recommendation for new users and the inability of recom- in hybrid mechanisms and their pros and cons.
mending the new products to users. Furthermore, the authors Table 7 illustrates the proposed methods and the compari-
of the paper also suggested issues such as considering more son of the most significant pros and cons of the articles. This
details of products like categories, price, description for use method works better than the previous ones. In other words,
in CBF algorithms, and other hybridization like monolithic in this method, given the combination of several methods,
hybrid and pipelined hybrid for future studies. the results are more accurate. However, it requires more
Saini et al. [100] proposed an approach to detect the processing times and higher computational costs.
sequences coming after clients while buying products to
develop the RS effectiveness. Since the purchase of some VI. RESULTS AND COMPARISON
products frequently takes place step by step, the idea of using We categorized and mentioned some of the RS methods in the
sequence pattern mining provided by the study. Contrary to previous section. We reviewed five basic techniques, includ-
other methods that consider customer profile and product ing CBF, CF, DBF, KBF, and hybrid, with expressing the
description, the authors not taken into account the users who advantages and disadvantages for each model. Also, we com-
not made any purchases, which is one of the weaknesses of pared the selected paper for such challenges and issues as
this approach. The proposed method improved the RSs by cold start, scalability, response time, data sparsity, accuracy,
suggesting various items that are carried one by one after performance, and security.
some months. Also, they evaluated the accuracy of the system Table 8 demonstrates an overview of the analyzed tech-
with testing data and showed a big advantage. However, niques of the RS and their fundamental challenges and
they proposed the future work to discover sequences for the issues like answer time, protection, precision, scalability,
particular client or the same client by employing a similar diversity/novelty/serendipity, operation cost, implicit/explicit
approach and include the sequences that come after the client data source, and independence. These features obtained from
in many years. selected papers and scored by a checkmark (X), cross (×),
As most e-commerce platforms and search engines and minus (-) sings. The minus and checkmark signs indicate
do not contain sentiment and semantic offering features, the least and the most incidents of these aspects, respectively.
Shaikh, et al. [134] proposed a system that considers seman- Moreover, Checkmark sign shows that the authors expressed
tic factor and sentiment in RS to get an improvement the metric as the advantage of the proposed method. On the
in their recommendations, and they proved their opinion. other hand, the cross sign represents the disadvantage of
They applied the graph algorithm, and the tags attached the feature. The minus sign explains that the authors not
product images with the assistance of overlap formula to considered the issue in the selected paper. A number of

115708 VOLUME 8, 2020

P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

TABLE 7. Comparing the most significant pros and cons of the hybrid mechanisms of selected papers in recommender systems in e-commerce.

the mentioned aspects are stated implicitly. Accordingly, in general, each method may consider several metrics we
the investigators determined the implicit aspects by content studied and ignore others due to the nature of the research. We
analysis of the chosen papers and scored them. suggest that researchers set their goal to balance the essential
In the Diversity/Novelty/Serendipity column, the letters metrics in their innovative methods because the efficiency of
‘‘D,’’ ‘‘N,’’ and ‘‘S’’ are used instead of diversity, novelty, new methods is measured based on the results of various met-
and serendipity, respectively. If the method does not any rics. Also, our selected metrics may be useful in all systems
of these features, a minus sign is used. In the next col- and do not specialize in recommender systems. For example,
umn, the letter ‘‘I’’ indicates the implicitness of the data, accuracy is the most critical metric for any information envi-
and the letter ‘‘E’’ indicates the explicit type of data used ronment, especially in recommender systems.
for the proposed method. If we show the ‘‘IE’’; it means that The results of the given comparison in Table 8 pointed
both data types are used by the method. Finally, in the last out that the authors of the selected papers have focused on
column, the letter ‘‘I’’ shows the independence of the imple- accuracy, response time, and scalability. Since many articles
mentation, like RA; otherwise, the letter ‘‘D’’ means that it did not meet our search criteria, we do not recommend to
should be used in the current e-commerce platform. Both generalize the conclusion. For instance, security is a vital
letters ‘‘DI’’ are for the proposed method that can be used in metric in many systems, but we resulted in many of the
both cases. selected papers have ignored it.
It is important to note that the results obtained in this review Figure 6 shows a stacked column diagram of the metrics
study were conducted only by comparing selected articles we reviewed in the selected papers.
and cannot be generalized to all methods of recommender Figure 7 shows a bar chart of the distribution of diver-
systems. For example, an article may have been intended sity, novelty, and serendipity of recommended items in
to increase scalability and ignore security issues. However, e-commerce methods in the selected papers. According to

VOLUME 8, 2020 115709

 P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

TABLE 8. An overview of the discussed RSS techniques and their main features.

Figure 7, diversity/novelty is the most considerable approach algorithms in the selected papers. Because of using these
to the selected papers’ authors. factors by recommender systems in e-commerce, the user’s
Figure 8 shows a bar chart of the distribution of fatigue disappears from repetitive and known recommenda-
implicit/explicit/both data sources for using e-commerce tions in the system. Also, it can be valuable in improving the

115710 VOLUME 8, 2020

P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

FIGURE 8. The bar chart of the distribution of using implicit or explicit

data in the reviewed papers.

FIGURE 6. The stacked column diagram of the metrics based on

comparative analysis.

FIGURE 9. The bar chart of the distribution of Independent feature in the

reviewed papers.

open problems of the matters discussed for future studies.

It is undeniable that no technique includes all issues relat-
ing to recommender systems in e-commerce. For example,
some methods provide scalability or response time, while
FIGURE 7. The bar chart of the distribution of using the diversity/
novelty/ serendipity in the reviewed papers.
some ignore these concerns. Also, some algorithms used the
computer simulation plan while some others use evaluation
frameworks. Such methods can be examined in a real-world
level of customer loyalty. We propose these essential topics e-commerce platform to offer efficient results. Furthermore,
for future studies. the authors of selected papers addressed some significant
Figure 9 shows a bar chart of the distribution of the inde- ideas of the discussed algorithms to be investigated in future
pendence feature of selected papers. According to Figure 9, studies. The personalized and context-sensitive recommenda-
dependence is the most considerable approach to the selected tions to mobile users are very interesting for future research
papers’ authors. In other words, most papers studied methods because of the complexity of mobile data, data distribution
that are independent of the current e-commerce system and in some fast-changing of the environment related to the
connect the system to operate. Thus, the independence prop- e-commerce platform, data sparsity problems, big data era,
erty of the RS method needs to research in future studies, and numerous dimensions of data [34].
especially in DBF and Hybrid methods, when issues such To enhance the quality of recommendation in recom-
as network overhead and information security are also of mender systems in e-commerce, many researchers suggested
paramount importance at the same time. various algorithms, especially for CF mechanisms. The CF
methods face some challenges such as sparsity, scalability,
VII. OPEN ISSUES and cold-start problem. Thus, even though e-commerce web-
As far as RSs are concerned, many vital issues are outside sites widely use recommendation techniques to improve their
the scope of the current research. This section covers the business, Sivapalan et al. [116] expressed that they still face

VOLUME 8, 2020 115711

 P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

some research and practical challenges, including scalability, Considering data changes and users’ preference changes
rich data, consumer-centered recommendations, anonymous in such systems is essential. Changes in the level of con-
users, and connecting recommenders to markets. sciousness in the community cause a change in the user’s
Suitable data types for use in recommender systems in temperament, so the recommender system should consider
e-commerce include on-site user activity (clicks, searches, these changes to process the results following the user’s
page, and item views), and off-site user activity (tracking desire.
clicks in the email and mobile applications), particular items Getting a complete picture of the user is so challenging
or user profiles data [135], and contextual data (the device because it needs to use and analyzing more data resources
used, current user location, and referral URL). On the other about the user. Therefore, it is necessary to make more
hand, the RSs requires sufficient data to run their algorithms relevant recommendations by joining different data sources
more carefully and provide the relevant recommendations. and possibly using user activities on other websites or social
Big data makes this possible [136]. However, its signifi- networks. It is also necessary that the results do not fatigue
cant problem is the high consumption of CPU and memory. the user, so it is vital to consider diversity, novelty, and
To solve the problem, the researchers proposed some solu- serendipity in the recommendations for each user.
tions such as using Hadoop and Mahout or some tools such The technical issues associated with algorithms used in
as PredictionIO. The traditional methods for processing this RSs, such as complexity and optimization, distributed meth-
large volume of data are not suitable at high speed. Therefore, ods, and creative approaches, are also suggested as future
the researchers should develop new strategies for using e- studies.
commerce data generated, especially on globally-developed
Web sites.
On the other hand, some studies do not consider all men- VIII. CONCLUSION AND LIMITATION
tioned metrics in section 6 to improve the proposed systems. In this paper, the researchers exposed the systematic liter-
For example, many researchers focused on their methods, and ature review of the latest issues in recommender systems
they did not mention security. So security gets the minimum in e-commerce. According to the performed SLR in recom-
focused against all other evaluation metrics in the selected mender systems in e-commerce mechanisms from 2008 to
papers. In the real-word, the performance of the recom- 2019, the number of published papers are discovered very
mender systems in e-commerce depends on multi-criteria. high in 2016. Additionally, famous publishers published most
For instance, Aznoli and Navimipour [32] pointed to some of the selected papers. We classify 33 preferred studies in
limitations of all RSs, such as the lack of data and a large five categories that 11 of them are CF techniques, three of
amount of data, changing data, changing users’ preferences, them are DBF techniques, six of them are KBF, five of them
unpredictable items, and complexity. are CBF, and eight of them are Hybrid techniques. We sum-
In content-centric techniques, problems like overspecial- marized the selected papers and showed the comparison
ization, limited-content analysis problems, and new user and evaluation of each proposed mechanism with significant
problems may occur. advantages and disadvantages to point out the essential issues
Thus, in summary, the most important challenges in RSs in for future researches.
e-commerce are the limitations related to big-data problems Furthermore, all selected mechanisms are compared based
or scalability challenges [121], Internet of things and person- on some crucial metrics such as security, response time, scala-
alized RSs, shared clients or devices, unknown user problem, bility, accuracy, operation cost, diversity/novelty/serendipity,
and multilingual and contextual processing matters. Based on implicit/explicit data source, and independence. The results
the previous section results, we found some open issues such confirmed that most of the studies work to improve the
as operation cost, optimization, and security in the field of accuracy of recommendations, but security, response time,
RSs in e-commerce [137]. novelty, diversity, serendipity are not considered in many
Nevertheless, despite extensive and useful research in the papers. In this study, we found that collaborating filter-
field of recommender systems in e-commerce, there are still ing techniques were used more than all other methods.
some issues that need to be considered for future researches. However, the latest papers focused on new approaches
Accordingly, the research is required to consider the fol- such as data and web mining algorithms, neural network-
lowing RSs limitations in the field of research and propose based methods, and genetic algorithms. We expressed the
relevant solutions: most critical open issues related to recommender systems
Both a lack of data and a significant amount of data cause in e-commerce.
problems in recommender systems in e-commerce. There Research limitations generally result from the choice of
are reasons to suggest that RSs result in useful recommen- papers, the selection method of relevant issues, used tech-
dations through the consideration of the adequacy of data. niques of analysis, and interpretation of review. We attempted
On the other hand, a vast amount of data consumes more to perform the systematic study of RSs used in e-commerce
computational resources and increases the response time for carefully, but it might some limitations. In this regard, future
the user. Therefore, future researches need to consider both studies must take into account the limitations of the current
issues. research as follows:

115712 VOLUME 8, 2020

P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

• Research Scope: Although the subject reviewed and [16] D. Galeano and A. Paccanaro, ‘‘A recommender system approach for pre-
documented in many scientific documents, such as aca- dicting drug side effects,’’ in Proc. Int. Joint Conf. Neural Netw. (IJCNN),
Jul. 2018, pp. 1–8.
demic publications, technical reports, editorial notes, [17] J. Chen, B. Wang, L. U, and Z. Ouyang, ‘‘Personal recommender system
and web pages, books, and papers, we limited just based on user interest community in social network model,’’ Phys. A, Stat.
academic leading international journals to obtain the Mech. Appl., vol. 526, Jul. 2019, Art. no. 120961.
[18] L. Liu, S. Yu, X. Wei, and Z. Ning, ‘‘An improved Apriori–based algorithm
best qualification. We excluded the papers published in for friends recommendation in microblog,’’ Int. J. Commun. Syst., vol. 31,
national journals and conferences. no. 2, p. e3453, Jan. 2018.
• Study and publication bias: We decided Google [19] L. Yao, Z. Xu, X. Zhou, and B. Lev, ‘‘Synergies between association rules
and collaborative filtering in recommender system: An application to auto
scholar as a reliable source for our search. However, industry,’’ in Data Science and Digital Business, F. P. G. Márquez and
there are all restrictions on Google Scholar in the present B. Lev, Eds. Cham, Switzerland: Springer, 2019, pp. 65–80.
study. So, we do not claim that we found and reviewed [20] J. K. Tarus, Z. Niu, and D. Kalui, ‘‘A hybrid recommender system for e-
learning based on context awareness and sequential pattern mining,’’ Soft
all of the related items. Comput., vol. 22, no. 8, pp. 2449–2461, Apr. 2018.
• Study queries: We preferred some keywords to select [21] R. Katarya and O. P. Verma, ‘‘Efficient music recommender system using
the relevant papers and search those in the title of papers. context graph and particle swarm,’’ Multimedia Tools Appl., vol. 77, no. 2,
pp. 2673–2687, Jan. 2018.
There may be research in which the research title does [22] J.-H. Park, ‘‘Resource recommender system based on psychological user
not include the different combinations of our selected type indicator,’’ J. Ambient Intell. Hum. Comput., vol. 10, no. 1, pp. 27–39,
keywords, but the article is entirely relevant to the sub- Jan. 2019.
[23] I. Mashal, O. Alsaryrah, and T.-Y. Chung, ‘‘Testing and evaluating recom-
ject. Therefore, we did not recognize such items. mendation algorithms in Internet of Things,’’ J. Ambient Intell. Humanized
• Classification: We ranked the selected papers in some Comput., vol. 7, no. 6, pp. 889–900, Dec. 2016.
categories, but it can be arranged creatively, too. [24] R. Y. Toledo, A. A. Alzahrani, and L. Martinez, ‘‘A food recommender
system considering nutritional information and user preferences,’’ IEEE
Access, vol. 7, pp. 96695–96711, 2019.
REFERENCES [25] S. JothiLakshmi and M. Thangaraj, ‘‘Design and development of recom-
mender system for target marketing of higher education institution using
[1] E. Turban, D. King, J. Lee, and D. Viehland, Electronic Commerce: A EDM,’’ Int. J. Appl. Eng. Res., vol. 13, no. 19, pp. 14431–14437, 2018.
Managerial Perspective. Upper Saddle River, NJ, USA: Prentice-Hall, [26] Y. Li, J. Liu, and J. Ren, ‘‘A novel learning model based on trust dif-
2002. fusion and global item for recommender systems,’’ IEEE Access, vol. 7,
[2] R. T. Wigand, ‘‘Electronic commerce: Definition, theory, and context,’’ Inf. pp. 170270–170281, 2019.
Soc., vol. 13, no. 1, pp. 1–16, Mar. 1997. [27] B. M. Shoja and N. Tabrizi, ‘‘Customer reviews analysis with deep neural
[3] P. Ratnasingham, ‘‘The importance of trust in electronic commerce,’’ networks for e-commerce recommender systems,’’ IEEE Access, vol. 7,
Internet Res., vol. 8, no. 4, pp. 313–321, Oct. 1998. pp. 119121–119130, 2019.
[4] Y. Guo, M. Wang, and X. Li, ‘‘Application of an improved Apriori algo- [28] J. B. Schafer, J. Konstan, and J. Riedi, ‘‘Recommender systems in e-
rithm in a mobile e-commerce recommendation system,’’ Ind. Manage. commerce,’’ in Proc. 1st ACM Conf. Electron. Commerce (EC), 1999,
Data Syst., vol. 117, no. 2, pp. 287–303, Mar. 2017. pp. 158–166.
[5] B. Zareie and N. J. Navimipour, ‘‘The effect of electronic learning systems [29] F. Ricci, L. Rokach, and B. Shapira, ‘‘Introduction to recommender sys-
on the employee’s commitment,’’ Int. J. Manage. Edu., vol. 14, no. 2, tems handbook,’’ in Recommender Systems Handbook, F. Ricci, L. Rokach,
pp. 167–175, 2016. B. Shapira, and P. B. Kantor, Eds. Boston, MA, USA: Springer, 2011,
[6] N. J. Navimipour and Z. Soltani, ‘‘The impact of cost, technology accep- pp. 1–35.
tance and employees’ satisfaction on the effectiveness of the electronic cus- [30] J. Wu, S. Guo, H. Huang, W. Liu, and Y. Xiang, ‘‘Information and com-
tomer relationship management systems,’’ Comput. Hum. Behav., vol. 55, munications technologies for sustainable development goals: State-of-the-
pp. 1052–1066, Feb. 2016. art, needs and perspectives,’’ IEEE Commun. Surveys Tuts., vol. 20, no. 3,
[7] S. H. Sharif, S. Mahmazi, N. J. Navimipour, and B. F. Aghdam, ‘‘A review pp. 2389–2406, 3rd Quart., 2018.
on search and discovery mechanisms in social networks,’’ Int. J. Inf. Eng. [31] Y. Charband and N. Jafari Navimipour, ‘‘Online knowledge sharing mech-
Electron. Bus., vol. 5, no. 6, p. 64, 2013. anisms: A systematic review of the state of the art literature and rec-
[8] N. J. Navimipour, A. M. Rahmani, A. H. Navin, and M. Hosseinzadeh, ommendations for future research,’’ Inf. Syst. Frontiers, vol. 18, no. 6,
‘‘Expert cloud: A cloud-based framework to share the knowledge and skills pp. 1131–1151, Dec. 2016.
of human resources,’’ Comput. Hum. Behav., vol. 46, pp. 57–74, May 2015. [32] F. Aznoli and N. J. Navimipour, ‘‘Cloud services recommendation:
[9] S. Hazratzadeh and N. J. Navimipour, ‘‘Colleague recommender system Reviewing the recent advances and suggesting the future research direc-
in the expert cloud using features matrix,’’ Kybernetes, vol. 45, no. 9, tions,’’ J. Netw. Comput. Appl., vol. 77, pp. 73–86, Jan. 2017.
pp. 1342–1357, Oct. 2016. [33] B. Xiao and I. Benbasat, ‘‘Research on the use, characteristics, and
impact of e-commerce product recommendation agents: A review and
[10] M. Ashouraie and N. J. Navimipour, ‘‘Priority-based task scheduling on
update for 2007–2012,’’ in Handbook Strategic e-Business Management,
heterogeneous resources in the expert cloud,’’ Kybernetes, vol. 44, no. 10,
F. J. Martínez-López, Ed. Berlin, Germany: Springer, 2014, pp. 403–431.
pp. 1455–1471, Nov. 2015.
[34] J. Lu, D. Wu, M. Mao, W. Wang, and G. Zhang, ‘‘Recommender sys-
[11] N. J. Navimipour, A. M. Rahmani, A. H. Navin, and M. Hosseinzadeh, tem application developments: A survey,’’ Decis. Support Syst., vol. 74,
‘‘Job scheduling in the expert cloud based on genetic algorithms,’’ Kyber- pp. 12–32, Jun. 2015.
netes, vol. 43, no. 8, pp. 1262–1275, Aug. 2014. [35] Y. Kilani, B. Alhijawi, and A. Alsarhan, ‘‘Using artificial intelligence
[12] A. Souri, A. M. Rahmani, and N. J. Navimipour, ‘‘Formal verification techniques in collaborative filtering recommender systems: Survey,’’ Int.
approaches in the Web service composition: A comprehensive analysis of J. Adv. Intell. Paradigms, vol. 11, nos. 3–4, pp. 378–396, 2018.
the current challenges for future research,’’ Int. J. Commun. Syst., vol. 31, [36] F. Karimova, ‘‘A survey of e-commerce recommender systems,’’ Eur. Sci.
no. 17, p. e3808, Nov. 2018. J., vol. 12, no. 34, p. 75, Dec. 2016.
[13] A. Simovic, ‘‘A big data smart library recommender system for an educa- [37] S. S. Li and E. Karahanna, ‘‘Online recommendation systems in a B2C
tional institution,’’ Library Hi Tech, vol. 36, no. 3, pp. 498–523, Sep. 2018. e-commerce context: A review and future directions,’’ J. Assoc. Inf. Syst.,
[14] P. Resnick and H. R. Varian, ‘‘Recommender systems,’’ Commun. ACM, vol. 16, no. 2, p. 72, 2015.
vol. 40, no. 3, pp. 56–58, 1997. [38] G. Adomavicius and A. Tuzhilin, ‘‘Toward the next generation of rec-
[15] S. Pal and S. Pal, ‘‘Recommender system in e-commerce: Bringing revo- ommender systems: A survey of the state-of-the-art and possible exten-
lution in marine product industry,’’ Indian J. Comput. Sci., vol. 3, no. 5, sions,’’ IEEE Trans. Knowl. Data Eng., vol. 17, no. 6, pp. 734–749,
pp. 7–13, 2018. Jun. 2005.

VOLUME 8, 2020 115713

 P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

[39] M. Aprilianti, R. Mahendra, and I. Budi, ‘‘Implementation of weighted [63] M. Gueye, T. Abdessalem, and H. Naacke, ‘‘Dynamic recommender sys-
parallel hybrid recommender systems for e-commerce in Indonesia,’’ tem: Using cluster-based biases to improve the accuracy of the predic-
in Proc. Int. Conf. Adv. Comput. Sci. Inf. Syst. (ICACSIS), Oct. 2016, tions,’’ in Advances in Knowledge Discovery and Management, vol. 5,
pp. 321–326. F. Guillet, B. Pinaud, G. Venturini, and D. A. Zighed, Eds. Cham,
[40] E. Sezgin and S. Özkan, ‘‘A systematic literature review on health rec- Switzerland: Springer, 2016, pp. 79–104.
ommender systems,’’ in Proc. E-Health Bioeng. Conf. (EHB), Nov. 2013, [64] A. B. Bondi, ‘‘Characteristics of scalability and their impact on perfor-
pp. 1–4. mance,’’ in Proc. 2nd Int. Workshop Softw. Perform., Ottawa, ON, Canada,
[41] M. M. Khan, R. Ibrahim, and I. Ghani, ‘‘Cross domain recommender 2000, pp. 195–203.
systems: A systematic literature review,’’ ACM Comput. Surv., vol. 50, [65] M. Li, Y. Xiang, B. Zhang, F. Wei, and Q. Song, ‘‘A novel organizing
no. 3, pp. 1–34, Oct. 2017. scheme of single topic user group based on trust chain model in social
[42] E. Çano and M. Morisio, ‘‘Hybrid recommender systems: A system- network,’’ Int. J. Commun. Syst., vol. 31, no. 1, p. e3387, Jan. 2018.
atic literature review,’’ Intell. Data Anal., vol. 21, no. 6, pp. 1487–1524, [66] J. Wang and Q. Tang, ‘‘Recommender systems and their security con-
Nov. 2017. cerns,’’ IACR Cryptol. ePrint Arch., Tech. Rep. 2015/1108, 2015. [Online].
[43] M. Subramani and E. Walden, ‘‘The impact of e-commerce announcements Available: http://eprint.iacr.org/
on the market value of firms,’’ Inf. Syst. Res., vol. 12, no. 2, pp. 135–154, [67] Y.-L. Wang, W.-Z. Tang, X.-J. Yang, Y.-J. Wu, and F.-J. Chen, ‘‘An effi-
Jun. 2001. cient method for autoencoder-based collaborative filtering,’’ Concurrency
[44] T. K. Shih, C. Chiu, H. Hsu, and F. Lin, ‘‘An integrated framework Comput., Pract. Exp., vol. 31, no. 23, p. e4507, 2018.
for recommendation systems in e-commerce,’’ Ind. Manage. Data Syst., [68] B. Kitchenham and S. Charters, ‘‘Guidelines for performing systematic
vol. 102, no. 8, pp. 417–431, Nov. 2002. literature reviews in software engineering,’’ EBSE, Keele, U.K., Tech. Rep.
[45] C.-M. Kwon and S.-Y. Kim, ‘‘Simulation study on e-commerce recom- EBSE-2007-01, 2007.
mender system based on a customer-product purchase-matrix,’’ in Proc. [69] F. Aznoli and N. J. Navimipour, ‘‘Deployment strategies in the wire-
AsiaSim, 2007, pp. 327–336. less sensor networks: Systematic literature review, classification, and
[46] Y.-M. Li, C.-T. Wu, and C.-Y. Lai, ‘‘A social recommender mechanism current trends,’’ Wireless Pers. Commun., vol. 95, no. 2, pp. 819–846,
for e-commerce: Combining similarity, trust, and relationship,’’ Decis. Jul. 2017.
Support Syst., vol. 55, no. 3, pp. 740–752, Jun. 2013. [70] A. Vakili and N. J. Navimipour, ‘‘Comprehensive and systematic review of
[47] D. Rosaci and G. M. L. Sarné, ‘‘REBECCA: A trust-based filtering the service composition mechanisms in the cloud environments,’’ J. Netw.
to improve recommendations for B2C e-commerce,’’ in Proc. AsiaSim. Comput. Appl., vol. 81, pp. 24–36, Mar. 2017.
Cham, Switzerland: Springer, 2007, pp. 31–36. [71] J. Wu, L. Ping, H. Wang, Z. Lin, and Q. Zhang, ‘‘Research on improved col-
[48] L. Jiang, Y. Cheng, L. Yang, J. Li, H. Yan, and X. Wang, ‘‘A trust- laborative filtering-based mobile e-commerce personalized recommender
based collaborative filtering algorithm for e-commerce recommendation system,’’ in Proc. Int. Conf. MultiMedia Inf. Technol. (MMIT), Dec. 2008,
system,’’ J. Ambient Intell. Hum. Comput., vol. 10, no. 8, pp. 3023–3034, pp. 143–146.
2019.
[72] Y. K. Choi and S. K. Kim, ‘‘An auxiliary recommendation system for
[49] Z. Hui-Hong, L. Yi-Jun, W.-Q. Zhang, and X. Jun-Yuan, ‘‘A survey of
repetitively purchasing items in e-commerce,’’ in Proc. Int. Conf. Big Data
recommender system applied in e-commerce,’’ Appl. Res. Comput., vol. 1,
Smart Comput. (BIGCOMP), Jan. 2014, pp. 96–98.
p. 3, Jan. 2004.
[73] C. Sun, R. Gao, and H. Xi, ‘‘Big data based retail recommender system
[50] J. S. Breese, D. Heckerman, and C. Kadie, ‘‘Empirical analysis of pre-
of non e-commerce,’’ in Proc. 5th Int. Conf. Comput., Commun. Netw.
dictive algorithms for collaborative filtering,’’ presented at the 14th Conf.
Technol. (ICCCNT), Jul. 2014, pp. 1–7.
Uncertainty Artif. Intell., Madison, WI, USA, 1998.
[74] S. Lin and X. Wenzheng, ‘‘E-commerce personalized recommendation
[51] U. Shardanand and P. Maes, ‘‘Social information filtering: Algorithms
system based on Web mining technology design and implementation,’’ in
for automating ‘word of mouth,’’’ presented at the SIGCHI Conf. Hum.
Proc. Int. Conf. Intell. Transp., Big Data Smart City (ICITBS), Dec. 2015,
Factors Comput. Syst., Denver, CO, USA, 1995.
pp. 347–350.
[52] J. Bennett, C. Elkan, B. Liu, P. Smyth, and D. Tikk, ‘‘KDD cup and
workshop 2007,’’ ACM SIGKDD Explor. Newslett., vol. 9, no. 2, pp. 51–52, [75] P. H. Aditya, I. Budi, and Q. Munajat, ‘‘A comparative analysis of memory-
Dec. 2007. based and model-based collaborative filtering on the implementation of
[53] Q. Le and A. Smola, ‘‘Direct optimization of ranking measures,’’ 2007, recommender system for e-commerce in Indonesia: A case study PT X,’’
arXiv:0704.3359. [Online]. Available: https://arxiv.org/abs/0704.3359 in Proc. Int. Conf. Adv. Comput. Sci. Inf. Syst. (ICACSIS), Oct. 2016,
[54] P. Cremonesi, Y. Koren, and R. Turrin, ‘‘Performance of recommender pp. 303–308.
algorithms on top-n recommendation tasks,’’ presented at the 4th ACM [76] J. Cao, H. Hu, T. Luo, J. Wang, M. Huang, K. Wang, Z. Wu, and
Conf. Recommender Syst., Barcelona, Spain, 2010. X. Zhang, ‘‘Distributed design and implementation of SVD++ algorithm
[55] C. D. Manning, P. Raghavan, and H. Schutze, Introduction to Infor- for e-commerce personalized recommender system,’’ in Proc. Nat. Conf.
mation Retrieval. Cambridge, U.K.: Cambridge Univ. Press, 2008, Embedded Syst. Technol. Springer, 2015, pp. 30–44.
p. 496. [77] G. F. Kuang, ‘‘The development of e-commerce recommendation system
[56] K. Järvelin and J. Kekäläinen, ‘‘Cumulated gain-based evaluation of based on collaborative filtering,’’ Adv. Eng. Forum, vol. 6, pp. 636–640,
IR techniques,’’ ACM Trans. Inf. Syst., vol. 20, no. 4, pp. 422–446, Sep. 2012.
Oct. 2002. [78] H. Hwangbo, Y. S. Kim, and K. J. Cha, ‘‘Recommendation system devel-
[57] C.-N. Ziegler, S. M. McNee, J. A. Konstan, and G. Lausen, ‘‘Improving opment for fashion retail e-commerce,’’ Electron. Commerce Res. Appl.,
recommendation lists through topic diversification,’’ presented at the 14th vol. 28, pp. 94–101, Mar. 2018.
Int. Conf. World Wide Web, Chiba, Japan, 2005. [79] L. Jiang, Y. Cheng, L. Yang, J. Li, H. Yan, and X. Wang, ‘‘A trust-based col-
[58] N. Lathia, S. Hailes, L. Capra, and X. Amatriain, ‘‘Temporal diversity in laborative filtering algorithm for e-commerce recommendation system,’’
recommender systems,’’ presented at the 33rd Int. ACM SIGIR Conf. Res. J. Ambient Intell. Hum. Comput., vol. 10, pp. 3023–3034, Jun. 2018.
Develop. Inf. Retr., Geneva, Switzerland, 2010. [80] R. S. Gaikwad, S. S. Udmale, and V. K. Sambhe, ‘‘E-commerce recom-
[59] A. H. Afridi, A. Yasar, and E. M. Shakshuki, ‘‘Facilitating research through mendation system using improved probabilistic model,’’ in Information
serendipity of recommendations,’’ J. Ambient Intell. Humanized Comput., and Communication Technology for Sustainable Development. Singapore:
vol. 11, no. 6, pp. 2263–2275, Jun. 2019. Springer, 2018, pp. 277–284.
[60] C.-N. Ziegler, S. M. McNee, J. A. Konstan, and G. Lausen, ‘‘Improving [81] Y. Sung Cho and K. Ho Ryu, ‘‘Implementation of personalized recommen-
recommendation lists through topic diversification,’’ in Proc. 14th Int. dation system using demograpic data and RFM method in e-commerce,’’
Conf. World Wide Web, 2005, pp. 22–32. in Proc. 4th IEEE Int. Conf. Manage. Innov. Technol. (ICMIT), Sep. 2008,
[61] D. Gonçalves, M. Costa, and F. Couto, ‘‘A flexible recommendation pp. 475–479.
system for cable TV,’’ 2016, arXiv:1609.02451. [Online]. Available: [82] Y. Fan, Y. Shen, and J. Mai, ‘‘Study of the model of e-commerce person-
https://arxiv.org/abs/1609.02451 alized recommendation system based on data mining,’’ in Proc. Int. Symp.
[62] K. Soliman, M. A. Mahmood, A. El Azab, and H. A. Hefny, ‘‘A survey Electron. Commerce Secur., 2008, pp. 647–651.
of recommender systems and geographical recommendation techniques,’’ [83] K. Souali, A. El Afia, and R. Faizi, ‘‘An automatic ethical-based recom-
in GIS Applications in the Tourism and Hospitality Industry. Hershey, PA, mender system for e-commerce,’’ in Proc. Int. Conf. Multimedia Comput.
USA: IGI Global, 2018, pp. 249–274. Syst. (ICMCS), Apr. 2011, pp. 1–4.

115714 VOLUME 8, 2020

P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

[84] L. Ya, ‘‘E-commerce personalized recommendation technology based on [106] V. Bajpai and Y. Yadav, ‘‘An improved dynamic e-commerce recommen-
expert system,’’ in Proc. 6th Int. Conf. Digit. Content, Multimedia Technol. dation system,’’ in Proc. Recent Adv. Interdiscipl. Trends Eng. Appl., 2019,
Appl. (IDC), Jun. 2010, pp. 342–345. pp. 1–11.
[85] M. I. Martín-Vicente, A. Gil-Solla, M. Ramos-Cabrer, [107] H. Jinming, ‘‘Application and research of collaborative filtering in e-
Y. Blanco-Fernández, and M. López-Nores, ‘‘Avoiding fake commerce recommendation system,’’ in Proc. 3rd Int. Conf. Comput. Sci.
neighborhoods in e-commerce collaborative recommender systems: Inf. Technol., vol. 4, Jul. 2010, pp. 686–689.
A semantic approach,’’ in Proc. 4th Int. Workshop Semantic Media [108] Y. Jiao and G. Cao, ‘‘A collaborative tagging system for person-
Adaptation Pers. (SMAP), 2009, pp. 9–14. alized recommendation in B2C electronic commerce,’’ in Proc. Int.
[86] Y.-C. Ku, C.-H. Peng, and Y.-C. Yang, ‘‘Consumer preferences for the Conf. Wireless Commun., Netw. Mobile Comput. (WiCom), Sep. 2007,
interface of e-commerce product recommendation system,’’ in HCI in pp. 3609–3612.
Business. Cham, Switzerland: Springer, 2014, pp. 526–537. [109] Y. Guo, X. Cheng, D. Dong, C. Luo, and R. Wang, ‘‘An improved collab-
[87] Y. Luo, ‘‘An analysis of Web mining-based recommender systems for e- orative filtering algorithm based on trust in e-commerce recommendation
commerce,’’ in Proc. 2nd Int. Conf. Comput. Appl. Syst. Model. Paris, systems,’’ in Proc. Int. Conf. Manage. Service Sci. (MASS), Aug. 2010,
France: Atlantis Press, 2012, pp. 167–170. pp. 1–4.
[88] W. Xiaosen, ‘‘The design and realization of personalized e-commerce [110] Y. Guo and G. Deng, ‘‘An improved personalized collaborative filterinng
recommendation system,’’ in Proc. Int. Conf. Social Sci. Technol. Educ. algolrithm in e-commerce recommender system,’’ in Proc. Int. Conf. Ser-
Paris, France: Atlantis Press, 2015, pp. 292–295. vice Syst. Service Manage., vol. 2, Oct. 2006, pp. 1582–1586.
[89] F. Huseynov, S. Y. Huseynov, and S. Özkan, ‘‘The influence of [111] J. Mai, Y. Fan, and Y. Shen, ‘‘A neural networks-based clustering col-
knowledge-based e-commerce product recommender agents on online laborative filtering algorithm in e-commerce recommendation system,’’ in
consumer decision-making,’’ Inf. Develop., vol. 32, no. 1, pp. 81–90, Proc. Int. Conf. Web Inf. Syst. Mining (WISM), Nov. 2009, pp. 616–619.
Jan. 2016. [112] Y. Fan, J. Mai, and X. Ren, ‘‘A rough set-based clustering collaborative
[90] P. Gatchalee, Z. Li, and T. Supnithi, ‘‘Ontology development for SMEs filtering algorithm in e-commerce recommendation system,’’ in Proc. Int.
e-commerce website based on content analysis and its recommendation Conf. Inf. Manage., Innov. Manage. Ind. Eng., vol. 4, 2009, pp. 401–404.
system,’’ in Proc. Int. Comput. Sci. Eng. Conf. (ICSEC), Sep. 2013, [113] A. M. Yağcı, T. Aytekin, and F. S. Gürgen, ‘‘Parallel pairwise learning
pp. 7–12. to rank for collaborative filtering,’’ Concurrency Comput., Pract. Exp.,
[91] Z. Gao, Z. Li, and K. Niu, ‘‘Solutions for problems of existing e-commerce vol. 31, no. 15, p. e5141, 2019.
recommendation system,’’ in Proc. IEEE Int. Conf. Consum. Electron.- [114] C. De Maio, G. Fenza, M. Gallo, V. Loia, and M. Parente, ‘‘Social
Taiwan (ICCE-TW), May 2016, pp. 1–2. media marketing through time-aware collaborative filtering,’’ Concurrency
[92] C. Chen, D. Wang, and Y. Ding, ‘‘User actions and timestamp based Comput., Pract. Exp., vol. 30, no. 1, p. e4098, Jan. 2018.
personalized recommendation for e-commerce system,’’ in Proc. IEEE Int. [115] H. Liu, X. Kong, X. Bai, W. Wang, T. M. Bekele, and F. Xia, ‘‘Context-
Conf. Comput. Inf. Technol. (CIT), Dec. 2016, pp. 183–189. based collaborative filtering for citation recommendation,’’ IEEE Access,
vol. 3, pp. 1695–1703, 2015.
[93] L. Palopoli, D. Rosaci, and G. M. Sarnè, ‘‘Introducing specialization in
e-commerce recommender systems,’’ Concurrent Eng., vol. 21, no. 3, [116] S. Sivapalan, A. Sadeghian, H. Rahnama, and A. M. Madni, ‘‘Recom-
pp. 187–196, Sep. 2013. mender systems in e-commerce,’’ in Proc. World Automat. Congr. (WAC),
2014, pp. 179–184.
[94] J. F. Miao, ‘‘Design and implementation of personalization recommen-
[117] S. K. Sharma and U. Suman, ‘‘Comparative study and analysis of Web
dation system in mobile e-commerce,’’ Adv. Mater. Res., vol. 989,
personalization frameworks of recommender systems for e-commerce,’’ in
pp. 4506–4509, Jul. 2014.
Proc. CUBE Int. Inf. Technol. Conf., 2012, pp. 629–634.
[95] Z. Qiu, M. Chen, and J. Huang, ‘‘Design of multi-mode e-commerce
[118] C. Haruechaiyasak, C. Tipnoe, S. Kongyoung, C. Damrongrat, and
recommendation system,’’ in Proc. 3rd Int. Symp. Intell. Inf. Technol. Secur.
N. Angkawattanawit, ‘‘A dynamic framework for maintaining customer
Informat., Apr. 2010, pp. 530–533.
profiles in e-commerce recommender systems,’’ in Proc. IEEE Int. Conf.
[96] F. Dong, J. Luo, X. Zhu, Y. Wang, and J. Shen, ‘‘A personalized e-Technol., e-Commerce e-Service, Jun. 2005, pp. 768–771.
hybrid recommendation system oriented to e-commerce mass data in the
[119] L.-T. Weng, Y. Xu, and Y. Li, ‘‘A framework for e-commerce oriented
cloud,’’ in Proc. IEEE Int. Conf. Syst., Man, Cybern. (SMC), Oct. 2013,
recommendation systems,’’ in Proc. Int. Conf. Act. Media Technol. (AMT),
pp. 1020–1025.
2005, pp. 309–314.
[97] W. Xue, B. Xiao, and L. Mu, ‘‘Intelligent mining on purchase information [120] C. Fu and Z. Leng, ‘‘A framework for recommender systems in e-
and recommendation system for e-commerce,’’ in Proc. IEEE Int. Conf. commerce based on distributed storage and data-mining,’’ in Proc. Int.
Ind. Eng. Eng. Manage. (IEEM), Dec. 2015, pp. 611–615. Conf. E-Bus. E-Government (ICEE), May 2010, pp. 3502–3505.
[98] M.-H. Jiao, X.-F. Chen, Z.-H. Su, and X. Chen, ‘‘Research on personalized [121] S. G. Walunj and K. Sadafale, ‘‘An online recommendation system for
recommendation optimization of e-commerce system based on customer e-commerce based on apache mahout framework,’’ in Proc. Annu. Conf.
trade behaviour data,’’ in Proc. Chin. Control Decis. Conf. (CCDC), Comput. People Res., 2013, pp. 153–158.
May 2016, pp. 6506–6511. [122] J. Sun and Y. Xie, ‘‘A Web data mining framework for e-commerce
[99] V. N. Zakharov and S. A. Philippov, ‘‘Clustering of goods and user profiles recommender systems,’’ in Proc. Int. Conf. Comput. Intell. Softw. Eng.
for personalizing in E-commerce recommender systems based on real (CiSE), Dec. 2009, pp. 1–4.
implicit data,’’ in Data Analytics and Management in Data Intensive [123] S. Sharma and A. Mahajan, ‘‘Suggestive approaches to create a recom-
Domains. Cham, Switzerland: Springer, 2017, pp. 178–191. mender system for GitHub,’’ Int. J. Inf. Technol. Comput. Sci., vol. 9, no. 8,
[100] S. Saini, S. Saumya, and J. P. Singh, ‘‘Sequential purchase recommenda- pp. 48–55, Aug. 2017.
tion system for E-commerce sites,’’ in Computer Information Systems and [124] S. M. Rezaeinia and R. Rahmani, ‘‘Recommender system based on
Industrial Management. Cham, Switzerland: Springer, 2017, pp. 366–375. customer segmentation (RSCS),’’ Kybernetes, vol. 45, no. 6, pp. 946–961,
[101] J. B. Schafer, D. Frankowski, J. Herlocker, and S. Sen, ‘‘Collaborative Jun. 2016.
filtering recommender systems,’’ in The Adaptive Web: Methods and [125] K. Raja and S. Pushpa, ‘‘Novelty-driven recommendation by using
Strategies of Web Personalization, P. Brusilovsky, A. Kobsa, and W. Nejdl, integrated matrix factorization and temporal-aware clustering opti-
Eds. Berlin, Germany: Springer, 2007, pp. 291–324. mization,’’ Int. J. Commun. Syst., p. e3851, Nov. 2018. [Online].
[102] P. Lops, M. de Gemmis, and G. Semeraro, ‘‘Content-based recommender Available: https://onlinelibrary.wiley.com/doi/abs/10.1002/dac.3851, doi:
systems: State of the art and trends,’’ in Recommender Systems Handbook, 10.1002/dac.3851.
F. Ricci, L. Rokach, B. Shapira, and P. B. Kantor, Eds. Boston, MA, USA: [126] H. J. Ahn, ‘‘A new similarity measure for collaborative filtering to allevi-
Springer, 2011, pp. 73–105. ate the new user cold-starting problem,’’ Inf. Sci., vol. 178, no. 1, pp. 37–51,
[103] S. Trewin, ‘‘Knowledge-based recommender systems,’’ Encyclopedia Jan. 2008.
Library Inf. Sci., vol. 69, no. 32, p. 180, 2000. [127] B. K. Patra, R. Launonen, V. Ollikainen, and S. Nandi, ‘‘A new similar-
[104] M. Y. H. Al-Shamri, ‘‘User profiling approaches for demographic recom- ity measure using Bhattacharyya coefficient for collaborative filtering in
mender systems,’’ Knowl.-Based Syst., vol. 100, pp. 175–187, May 2016. sparse data,’’ Knowl.-Based Syst., vol. 82, pp. 163–177, Jul. 2015.
[105] R. Burke, ‘‘Hybrid recommender systems: Survey and experiments,’’ [128] S. T. Yuan and C.-Y. Yang, ‘‘Service recommender system based on
User Model. User-Adapted Interact., vol. 12, no. 4, pp. 331–370, emotional features and social interactions,’’ Kybernetes, vol. 46, no. 2,
Nov. 2002. pp. 236–255, Feb. 2017.

VOLUME 8, 2020 115715

 P. Malekpour Alamdari et al.: Systematic Study on the RSs in the E-Commerce

[129] S. C. Cazella, E. Reategui, and L. O. C. Alvares, ‘‘E-commerce recom- MEHDI HOSSEINZADEH received the B.E.
menders’ authority: Applying the user’s opinion relevance in recommender degree in computer hardware engineering from
systems,’’ in Proc. 12th Brazilian Symp. Multimedia web, 2006, pp. 71–78. Islamic Azad University (IAU) at Dezful Branch,
[130] I.-G. Chun and I.-S. Hong, ‘‘The implementation of knowledge-based Iran, in 2003, and the M.Sc. and Ph.D. degrees in
recommender system for electronic commerce using java expert system computer system architecture from IAU Science
library,’’ in Proc. IEEE Int. Symp. Ind. Electron. (ISIE), vol. 3, Jun. 2001, and Research Branch, Tehran, Iran, in 2005 and
pp. 1766–1770. 2008, respectively. He is currently an Associate
[131] T. Tran, ‘‘Designing recommender systems for e-commerce: An inte-
Professor with the Iran University of Medical
gration approach,’’ in Proc. 8th Int. Conf. Electron. Commerce New E-
Sciences (IUMS), Tehran. His research interests
Commerce, Innov. Conquering Current Barriers, Obstacles Limitations
Conducting Successful Bus. Internet, 2006, pp. 512–518. include SDN, information technology, data min-
[132] A. Gogna and A. Majumdar, ‘‘A comprehensive recommender system ing, big data analytics, E-commerce, E-marketing, and social networks.
model: Improving accuracy for both warm and cold start users,’’ IEEE
Access, vol. 3, pp. 2803–2813, 2015.
[133] M. Mansoury and M. Shajari, ‘‘Improving recommender systems’ per-
formance on cold-start users and controversial items by a new similarity
model,’’ Int. J. Web Inf. Syst., vol. 12, no. 2, pp. 126–149, Jun. 2016.
[134] S. Shaikh, S. Rathi, and P. Janrao, ‘‘Recommendation system in e-
commerce Websites: A graph based approached,’’ in Proc. IEEE 7th Int.
Advance Comput. Conf. (IACC), Jan. 2017, pp. 931–934.
[135] S. Ouaftouh, A. Zellou, and A. Idri, ‘‘Social recommendation: A user
profile clustering-based approach,’’ Concurrency Comput., Pract. Exper.,
vol. 31, no. 20, p. e5330, 2019, doi: 10.1002/cpe.5330.
[136] H. Hu, Y. Wen, T.-S. Chua, and X. Li, ‘‘Toward scalable systems for big ALI ASGHAR SAFAEI received the B.Sc. degree
data analytics: A technology tutorial,’’ IEEE Access, vol. 2, pp. 652–687, in computer software engineering from Shahid
2014. Sattari Air University, Tehran, Iran, in 2001,
[137] Y. Charband and N. J. Navimipour, ‘‘Knowledge sharing mechanisms the M.Sc. degree in computer software engineer-
in the education: A systematic review of the state of the art literature ing from Ferdowsi University, Mashhad, Iran,
and recommendations for future research,’’ Kybernetes, vol. 47, no. 7, in 2004, and the Ph.D. degree in computer software
pp. 1456–1490, Aug. 2018. engineering from the Iran University of Science
and Technology, Tehran, in 2011. He is currently
an Assistant Professor of the Department of Com-
puter Engineering, Tarbiat Modares University,
PEGAH MALEKPOUR ALAMDARI received the Tehran. His research interests include medical software design, nursing
B.S. degree in computer and software engineer- informatics, free/open-source software (FOSS) development, advanced
ing from Islamic Azad University International databases, advanced software engineering, and distributed systems.
Jolfa Branch, Jolfa, Iran, in 2010, and the M.Sc.
degree in computer science from Tabriz Univer-
sity, Tabriz, Iran, in 2013. She is currently pur-
suing the Ph.D. degree with the Department of
Computer Engineering, Qeshm Branch, Islamic
Azad University, Qeshm, Iran. Her research inter-
ests include recommender systems, E-commerce,
machine learning, data mining, and programming.

ASO DARWESH received the B.Sc. degree in

NIMA JAFARI NAVIMIPOUR received the B.S. mathematics from the University of Sulaimani,
degree in computer and software engineering and Iraq, in 2001, the M.Sc. degree in computer
the M.S. degree in computer engineering and science from the University of Rene Descartes,
computer architecture from Islamic Azad Univer- in 2007, France, and the Ph.D. degree in computer
sity at Tabriz Branch, Tabriz, Iran, in 2007 and science from the University of Pierre and Mari
2009, respectively, and the Ph.D. degree in com- Curie, France, in 2010. He is currently an Asso-
puter engineering and computer architecture from ciate Professor with the Department of Computer
Islamic Azad University Science and Research Science, University of Human Development, Iraq.
Branch, Tehran, Iran, in 2014. His research inter- His research interests include serious games, adap-
ests include QCA, cloud computing, grid systems, tive learning, and cognitive diagnosis in E-learning.
computational intelligence, evolutionary computing, and wireless networks.

115716 VOLUME 8, 2020