Hybrid Recommendation Systems: A State of Art

Fatima Zohra Trabelsi, Amal Khtira and Bouchra El Asri

IMS Team, ADMIR Laboratory, Rabat IT Center, ENSIAS, Mohammed V University, Rabat, Morocco

Keywords:

Recommendation Systems, Hybrid Filtering, Collaborative Filtering, Content-based Filtering,

Recommendation Problems, State of Art.

Abstract:

Recommendation systems have become more important and popular in many application areas such as music,

movies, e-commerce, advertisement and social networks. Recommendation systems use either collaborative

ﬁltering, content-based ﬁltering or hybrid ﬁltering in order to propose items to users, and each type has its

weaknesses and strengths. In this paper, we present the results of a literature review that focuses speciﬁcally

on hybrid recommendation systems. The objective of this review is to identify the problems that hybrid

ﬁltering tends to solve and the different techniques used to this end.

1 INTRODUCTION

Recommendation systems are techniques that pro-

pose products and services that are likely to interest

the users of a platform, a website or an application.

Nowadays, the demand for recommendation systems

has increased in many areas such as movies, music,

news, e-commerce, advertisement, tourism and social

networks (Pandey and Rajpoot, 2016). Those sys-

tems are analyzing several characteristics of the on-

line users, such as their reviews and their purchase

history to make suggestions to other users based on

the assumption that users with similar proﬁles tend to

make the same choices (Tsolakidis et al., 2016).

In real life, when going for online products, peo-

ple ﬁrst always look for those items that are of their

preferences. For that reason, recommendation sys-

tems, by the virtue of their nature, are used as a rem-

edy that helps users choose between the immense va-

riety of items (services and products) that Websites

offer (Pandey and Rajpoot, 2016). Recommendation

systems usually needs to consider many factors, in-

cluding accuracy, diversity, novelty, coverage, user

satisfaction and so on, to provide satisfactory recom-

mendations for users (Cai et al., 2020). Typically,

there are three major recommendation methods: col-

laborative, content-based and hybrid (Prakash et al.,

2019).

Collaborative ﬁltering is the most widely-used

technique in recommendation systems because it is

a method that makes predictions about a given user’s

interests by collecting a number of other users’ appre-

ciations (Duzen and Aktas, 2016), and this method

can be either user-based recommendation or item-

based recommendation. Content-based recommen-

dation is also a very important type of ﬁltering, be-

cause it computes recommendations according to the

features of items that the user preferred in the past

(Do et al., 2020). Recommendation can be for books,

movies, music, news, articles, documents, etc. In or-

der to have better recommendations for users, a new

recommendation technique was proposed in the lit-

erature by combining both the collaborative ﬁltering

and the content-based ﬁltering, which helps to ben-

eﬁt from the advantages of each method and over-

come their drawbacks. This type of combination is

known as the hybrid recommendation and aims at

providing more accurate and effective recommenda-

tions by merging different algorithms (Ammar et al.,

2020)(Tian et al., 2019)(Mansur et al., 2017)(Pandey

and Rajpoot, 2016).

In this paper, we summarize the state of the art of

hybrid recommendation systems in the last ﬁve years.

For this, we have conducted a literature review whose

objective is to collect the different problems of rec-

ommendation addressed by hybrid approaches and to

identify the techniques used to solve these problems.

The remaining of this paper is organized as fol-

lows. Section 2 gives an overview of the background

of our work. Section 3 presents the research method-

ology followed in the review. Section 4 answers the

research questions by analysing and discussing the re-

sults of the review. Finally, Section 5 concludes the

paper.

Trabelsi, F., Khtira, A. and El Asri, B.

Hybrid Recommendation Systems: A State of Art.

DOI: 10.5220/0010452202810288

In Proceedings of the 16th International Conference on Evaluation of Novel Approaches to Software Engineering (ENASE 2021), pages 281-288

ISBN: 978-989-758-508-1

281

2 BACKGROUND

Because of the rapid development of social network,

recommendation systems appear to sort through and

ﬁnd what is desired by users (Prakash et al., 2019), in

order to satisfy their needs. The ﬁrst works in the ﬁeld

were based only on the two known types of recom-

mendation, which are content-based ﬁltering (Shah

et al., 2017) (Kumar et al., 2018) (Najmani et al.,

2019) and collaborative ﬁltering (Song et al., 2016)

(Patel et al., 2017) (Maihami et al., 2019). And since

these two techniques have their strengths and weak-

nesses, researchers started to shift to hybridization be-

tween the two ﬁltering techniques in order to exploit

the advantages of both of them (Dhawan, 2019).

2.1 Content-based Recommendation

Content-based ﬁltering methods (Burke, 2002) are

based on a description of the item and the proﬁle of

the user’s preferences. These algorithms try to recom-

mend items that are similar to those that a user liked in

the past (Tian et al., 2019)(Nikzad–Khasmakhi et al.,

2019). Indeed, a content-based recommendation sys-

tem tracks user preferences in terms of items con-

sumed and liked, and from that data, it creates a user

proﬁle. Then, the system matches items from users’

proﬁle to those other items in the database and rec-

ommend the items the user has not consumed in the

past and that are similar to user’s preferences (Wairegi

et al., 2020).

2.2 Collaborative Recommendation

Collaborative ﬁltering method (Wairegi et al., 2020)

is based on opinion shared between users and their

taste. We recommend to a user an item appreciated by

another with common interests (Ammar et al., 2020)

(Chen et al., 2018). The fact of being capable of mak-

ing predictions, without asking for more data from the

users or items, gives more utility and importance for

the ﬁltering. However, there are many problems re-

lated to this type of systems, namely sparsity, scala-

bility and the cold start problem (Duzen and Aktas,

2016). This ﬁltering is divided into two categories:

user-based and item-based ﬁltering (Burke, 2002).

In the user-based approach, the algorithm pro-

duces a rating for each item using similar users (Li

et al., 2018a). That means if User 1 is interested in

Items A, B and C, and User 2 is interested in Items A

and B, then the system recommends Item C to User

The item-based technique computes recommen-

dation using the similarity between items and not be-

tween users (Patel et al., 2017). In other words, if a

user likes the Items A, B and C, and Item D is similar

to C, the system proposes Item D to this user.

2.3 Hybrid Recommendation

Hybrid recommendation systems (Burke, 2002) can

produce outputs that outperform single component

systems by combining multiple techniques of differ-

ent types, such as mixing content-based and collab-

orative ﬁltering methods, which is the most common

combination (Wairegi et al., 2020). Furthermore, it is

also possible to mix different techniques of the same

type (Cano and Morisio, 2017).

The collaborative recommendation approach and

the one based on the content are considered as

complementary, because the collaborative recommen-

dation does only recommend items already evalu-

ated, meanwhile, the content-based recommendation

is able to recommend new items not evaluated yet by

the user.

In addition, the content of the item is either in-

adequate or difﬁcult to extract. There is no need for

the content in the collaborative recommendation; in

contrast, this is very necessary in the second type of

ﬁltering.

To reach the best performance, we need an ap-

proach that deals with the different weaknesses of

individual technique, beneﬁts from their advantages

and relies on multiple sources in order to use the

most appropriate ones. In this vein, the hybridisa-

tion of the two types of ﬁltering have been proposed.

There are many available hybridization techniques

such as mixing, switching, weighted approach, fea-

ture combination among others. By using these tech-

niques, the system combines various recommendation

approaches to a single hybrid recommending system

(Wairegi et al., 2020).

3 RESEARCH METHODOLOGY

Several studies related to the hybrid recommendation

systems have been proposed by researchers and prac-

titioners. To analyse these studies, we have conducted

a review of the literature by following the same pro-

tocol of a Systematic Literature Review (SLR) de-

scribed in Kitchenham’s guidelines (Kitchenham and

Charters, 2007). This protocol contains the following

steps : 1) Identiﬁcation of research questions, 2) Re-

search in Databases, 3) Data Selection which includes

the deﬁnition of Inclusion and Exclusion criteria, 4)

Data Extraction, and 5) Data Analysis. The rest of

this section focuses on the ﬁrst four steps, while the

ENASE 2021 - 16th International Conference on Evaluation of Novel Approaches to Software Engineering

282

data analysis step is detailed in Section 4.

3.1 Search Questions

The objective of our review is to ﬁnd the different

contributions proposed in literature in relation with

hybrid recommendation systems and to discuss the

different criteria of hybridisation used by these ap-

proaches. We thus formulated the three following

questions :

• RQ1. What are the different approaches proposed

regarding hybrid recommendation systems?

• RQ2. What are the types of problems that the hy-

bridization techniques tend to solve?

• RQ3. What are the different techniques of hy-

bridization used to enhance the functioning of rec-

ommendation systems?

In order to answer the research questions deﬁned

above, we have constructed the research string using

the keywords related to our topic. The basic key-

words are: Recommendation, System, Hybridization

and Filtering. To make the research more efﬁcient,

we deﬁned a set of synonyms and alternatives for the

different keywords. To link the alternative keywords,

we used the Boolean “OR” and to interconnect the

different parts of the string, we used the Boolean

“AND”. As a consequence, we obtained the research

string presented below:

(hybrid OR hybridization OR collaborative OR

content-based) AND (recommender OR recommend-

ing OR recommendation) AND (system OR systems

OR ﬁltering OR technique)

3.2 Research in Databases

Using the keywords above, we considered publica-

tions retrieved from IEEE Xplore, ACM Digital Li-

brary, ScienceDirect and Springer Link.

• IEEE Xplore: This database is very easy to ma-

nipulate. First, we entered the search string, and

then we ﬁltered the ﬁrst result by date to obtain

only the papers corresponding to our review.

• ACM Digital Library: In this database, we had

to be more speciﬁc in the notation by adding two

quotes to the keywords, otherwise, we come up

with a big number of studies that have nothing to

do with our topic.

• ScienceDirect – Elsevier: This database works in

the same way as the ACM Digital Library in using

the quotes. In ScienceDirect, we can reﬁne our

search through many ﬁlters like date, publication

title, article type, etc.

• Springer Link: Springer link is also a database

known for its diversity in studies. The only com-

plicated issue we had to deal with in this database

is the interconnection between some ﬁlters.

3.3 Data Selection

To perform a successful literature review, the inclu-

sion and exclusion criteria must be carefully deﬁned

in order to keep only the articles that are relevant to

our search.

Figure 1: Papers Selection Process.

The inclusion criteria we have deﬁned are the fol-

lowing:

• IC1: The paper is a full article, a book, a chapter,

a report, a thesis, a presentation.

• IC2: The title or the abstract of the paper contains

the keywords of the search.

• IC3: The paper addresses the hybrid recommen-

dation systems.

• IC4: The paper addresses at least one problem

of recommendation or proposes at least one

technique of hybridization.

And the exclusion criteria are the following:

• EC1: The publication date is previous to 2016.

• EC2: The paper is written in a language other

than English.

• EC3: The paper is a short article, a standard, a

poster, an editorial, or a tutorial.

• EC4: The title, the keywords and the abstract do

not correspond to the research subject.

• EC5: The paper does not discuss the hybrid rec-

ommendation systems.

Hybrid Recommendation Systems: A State of Art

283

The total number of papers initially retrieved is

1816, divided as follows: 1046 from IEEE Xplore,

431 from ACM Digital Library, 245 from ScienceDi-

rect and 94 from SpringerLink. After applying all the

exclusion criteria, we kept only 33 papers at the end

of the selection process.

3.4 Data Extraction

In order to make a synthesis of the data collected and

to be able to answer the predeﬁned research questions

above, we extracted a number of attributes from each

selected paper, as described in Table 1 (Kitchenham,

2007),

Table 1: Attributes used in data extraction.

Title Title of the paper

Year Publication year of the paper

Type e. g. Journal paper, conference pa-

per, thesis, book, chapter

Database e. g. IEEE, ACM, SpringerLink,

ScienceDirect (Elsevier)

Keywords Keywords speciﬁed in the paper

Methodology Methodology followed in the study

Contribution e. g. Model, Framework, Tools,

Method, Algorithm

4 RESULTS AND DISCUSSION

The objective of this step is to answer the research

questions deﬁned in the last section. The different

papers we have judged relevant for our review were

studied from different perspectives.

4.1 Metadata Analysis

In our review, we analyzed many types of articles

metadata, but we present in this article two main

types, which are the Data Source and the Publication

Year.

Data Source. Figure 2 presents the percentage

of papers from the four digital databases. The

distribution of papers is the following : 39% of the

selected papers belong to IEEE Explore (13 papers),

23% of papers from Springer (8 papers) and the

lowest percentage of papers were retrieved from both

ScienceDirect and ACM 19% (6 papers).

Figure 2: Percentage of papers in databases.

Publication Year. As mentioned earlier, the review

was conducted for the period 2016-2020. Figure 3

shows the number of papers published in each year.

The highest number of papers were published in 2019

with 12 papers. 7 of the selected papers were pub-

lished in 2017, 6 papers were published in 2020, 4

papers in 2018 and 4 papers in 2016.

Figure 3: Number of papers per year.

4.2 Problems Addressed by

Hybridisation

To answer the question RQ2 asked above, we summa-

rized the main problems that the selected papers try to

avoid by using the hybridization. A total of 12 prob-

lems were found, and each problem has a degree of

importance. This section deﬁned the four more com-

mon problems:

• Cold-start: This problem is caused when the sys-

tem is unable to recommend an item to any user

because there is no rating for that item, or when

a new user enters in the system and has no rating

record, the system is unable to recommend items

to him (Wang et al., 2019).

• Data Sparsity: This problem is caused by the

rate of users, which means if the user do not rate

some item, the system suffers from the paucity of

data and becomes unable to recommend items to

him because it has no idea about the user taste

(Dhawan, 2019).

ENASE 2021 - 16th International Conference on Evaluation of Novel Approaches to Software Engineering

284

Table 2: Problems Addressed per Paper.

Paper Cold-Start Sparsity Scalability Diversity Others

(Pandey and Rajpoot, 2016) X X X X X

(Dhawan, 2019) X X

(Patel et al., 2017) X X X

(Duzen and Aktas, 2016) X X

(Mansur et al., 2017) X

(Tian et al., 2019) X X X X X

(Nikzad–Khasmakhi et al., 2019) X X X X

(Idrissi et al., 2019a) X X X

(Wairegi et al., 2020) X X X

(Agner et al., 2020) X X

(Cai et al., 2020) X X

(Ammar et al., 2020) X X X X X

(Tsolakidis et al., 2016) X X X X X

(Kumar et al., 2018) X X X

(Cano and Morisio, 2017) X X X X X

(Tang and Wang, 2016) X X X X

(Prakash et al., 2019) X

(Gong et al., 2020) X X

(Cao et al., 2018) X X

(Zhang et al., 2018) X X X X

(Alamdari et al., 2017) X X X X X

(Chen et al., 2017) X X X

(Gulzar et al., 2018) X

(Do et al., 2020) X

(Chen et al., 2018) X

(Wang et al., 2019) X

(Li et al., 2018a) X X

(Maihami et al., 2019) X X

(Shah et al., 2017) X X X

(Song et al., 2016) X

(Najmani et al., 2019) X X X X X

(C¸ akır et al., 2019) X

(Idrissi et al., 2019b) X

• Scalability: It is the increase of number of rated

items by users, which increases also the complex-

ity. Thus, the recommendation system is unable

to handle such amount of data (Alamdari et al.,

2017).

• Diversity: Diversity is recommending the same

item many times that are presented with differ-

ent names but represent the same product or item.

In this case, the system is unable to identify the

item with the other name affected to it (Burke,

2002). Diversity is one of the optimization objec-

tives, and it is related to the accuracy (Cai et al.,

2020). Diversity of recommended items decreases

while accuracy improves.

There are other problems appearing in few stud-

ies, like accuracy, gray sheep, shilling attacks,

black sheep, Changing user preferences, privacy,

trust.

Figure 4 shows the total of papers that treat each

problem. It is very notable that the cold-start issue is

the main problem treated in this area, With 28 papers

out of the total of papers (85%). The second most ad-

dressed issue is data sparsity with 26 papers (79%).

Scalability, as discussed before, is also an interest-

ing issue in the ﬁeld of hybrid recommendation and

is covered by 17 papers selected in this review (52%).

And ﬁnally, some of these studies addressed data

diversity with 11 papers (33%), but, not all of them,

seeing that the over-specialization of data is still caus-

ing problem in some of type of hybridization.

The ﬁgure also shows that the other different is-

sues, mentioned separately in few papers, are pre-

sented in less than 25% out of 33 papers with 8 pa-

pers.

Table 2 presents the Problems addressed and the

different papers that deal with them.

Hybrid Recommendation Systems: A State of Art

285

Table 3: Problems vs. hybridisation techniques.

Cold-start Diversity Sparcity Scalability

Weighting YES YES YES YES

Switching YES NO YES YES

Mixture YES NO YES YES

Feature combination YES NO YES YES

Feature augmentation YES YES YES YES

Meta-level YES YES YES YES

Figure 4: Number of papers per Recommendation Problem.

4.3 Hybridisation Techniques

The quality of recommendation is the main topic of

all the papers. Quality has many deﬁnitions, but gen-

erally it is the ability of a product to satisfy deﬁned

requirements [18]. Based on the selected papers, we

collected the different hybridisation techniques used

to solve the recommendation problems.

• Weighting: In the weighted hybrid recommen-

dation, score or weight of a recommended item

is calculated from the results of all available rec-

ommendation techniques implemented in the sys-

tem (Burke, 2002). In this sense, (Li et al.,

2018a) have worked on the recommendation sys-

tem based on weighted linear regression models to

establish the model between the user’s scores for

the items and the user’s highest frequency scores.

• Switching: By using this technique, the system

switches between recommendation techniques de-

pending on the current situation (Burke, 2002).

This technique has been used by many papers. For

example, (Prakash et al., 2019) proposed a system

that optimizes the results by incorporating dual

hybridization techniques, meta-level and switch-

ing.

• Cascade: Cascade technique selects candidate

completely with the main recommendation, and

uses the other recommendation to reﬁne product

or item scores (Burke, 2002). An example of sys-

tems that use this technique is the mobile music

cascade recommender that combines SVM genre

classiﬁcation with collaborative user personality

(Cano and Morisio, 2017).

• Feature Combination: Features from different

recommendation data sources are thrown together

into a single recommendation algorithm (Burke,

2002). This technique is used by (Wairegi et al.,

2020) to come up with an hybrid recommender

system that combines both content-based and

collaborative ﬁltering approaches to recommend

items to the users.

• Mixture: Mixed hybrids combine recommen-

dation results of different recommendation tech-

niques at the same time instead of having just one

recommendation per item (Burke, 2002). In this

contexte, (Gulzar et al., 2018) proposed a system

based on a mixed combination of three individual

techniques used in recommender information re-

trieval systems.

• Feature Augmentation: The output from one

technique is used as an input feature to another

(Burke, 2002). For instance, (Cano and Morisio,

2017) mentioned an hybrid method that combines

multidimensional clustering and Collaborative ﬁl-

tering to increase recommendation diversity.

• Meta-level: The meta-level technique seeks to in-

put the result obtained by collaborative ﬁltering

into the content-based system to get a more re-

ﬁned recommendation set (Burke, 2002). Among

papers applying this technique, (Sattar et al.,

2017) proposed meta-level hybrid recommenda-

tion algorithms by combining item-based collab-

orative ﬁltering with content-based ﬁltering and

build content-based ﬁltering model on the content

of K-nearest neighbors of items.

4.4 Discussion

Table 3 presents the techniques used to solve the dif-

ferent recommendation problems addressed by the se-

lected papers. From the table, we can see clearly that

hybridization techniques in most papers had for ob-

jective to overcome the cold-start problem, data spar-

sity and scalability. The problem then is with the di-

versity of data or the problem of over-specialization of

ENASE 2021 - 16th International Conference on Evaluation of Novel Approaches to Software Engineering

286

data which is still existing in Switching Hybrid Rec-

ommenders. Indeed, (Ghazanfar and Prugel-Bennett,

2010a) talks about the four issues of a hybrid recom-

mender system, but the switching hybridization pro-

posed in the paper is only able to deal with with the

cold-start, sparsity and scalability problems. A good

quality is also saving the users’ valuable time by rec-

ommending the best items that are related to their

preferences and choices and a weighted system has

shown the ability to make consolidated decisions (Do

et al., 2020) and to overcome the problem of data

diversity or the over-specialization besides the cold-

start issue, data sparsity and scalability (Cano and

Morisio, 2017). As for mixed hybridization, it has

shown its ability to avoid cold-start, sparsity and scal-

ability issues (Santos, 2014) but seems to have still

the data over-specialization issue since multiple rec-

ommenders present their results at once.

Hybridization based on feature combination treats

information as simply additional feature data (Burke,

2002), which explains why the combination type is

facing the same problem of data diversity as the

weighted and switching types. In cascading hybrid

recommender systems, a recommendation technique

is applied to produce a coarse candidate list of items

for recommendation that are reﬁned by applying other

recommendation techniques (Ghazanfar and Prugel-

Bennett, 2010b). This means that the cascading type

does not suffer from the over-specialization issue due

to the enhancement that does on the other previous

techniques. Hybridization based on feature augmen-

tation is also able to deal with data over-specialization

issue (Li et al., 2018b), because it is employed to pro-

duce a rating or classiﬁcation of items before recom-

mending them to the users. Finally, the meta-level

technique gives a compressed representation of user’s

interest, which explains how it deals with diversity

problem and data sparsity too.

5 CONCLUSION

Today, recommender systems play an important role

in our daily life. It have become an essential tool for

users to navigate the vast number of options for con-

tent and products, because it enables users to make

the most appropriate choices from the immense vari-

ety of items that are available by predicting the pref-

erences that users would give to an item (Tang and

Wang, 2016).

In this papers, we have presented the results of a

review that focuses especially on hybrid recommen-

dation systems. The objective of this review was to

investigate the different approaches proposed in the

subject between 2016 and 2020. As a result of search-

ing papers in four digital libraries, we identiﬁed at the

beginning 1816 papers. By applying a set of inclu-

sion and exclusion criteria, 33 relevant papers were

selected.

Many conclusions have been drawn from this re-

view. The most important constraint addressed by hy-

brid approaches is the cold start problem because it

appears in almost 75% of papers.

The analysis has also shown that data sparsity is

another important issue in recommendation, while the

scalability problem comes in the third place. There

are other issues found in studies such as accuracy,

gray sheep, shilling attacks, black sheep, Changing

user preferences, privacy and trust, that are also im-

portant but do not have the same impact on recom-

mendation.

From the results, we can conclude that despite of

representing an important issue in recommendation

systems, data diversity or also known as the over-

specialization of information is not as current as the

other issues. It is slightly studied or just mentioned

in some papers. Therefore, we have found out that

this issue is not addressed by some hybridization tech-

niques, namely the weighting, the switching and the

feature combination techniques.

REFERENCES

Agner, L., Necyk, B., and Renzi, A. (2020). Recommen-

dation systems and machine learning: Mapping the

user experience. In Marcus A., Rosenzweig E. (eds)

Design, User Experience, and Usability. Design for

Contemporary Interactive Environments. Springer.

C¸ akır, M.,

u, . G., and Tugay, R. (2019). A deep hy-

brid model for recommendation systems. In Alviano

M. et al. (eds) AI*IA 2019 – Advances in Artiﬁcial In-

telligence. LNCS, vol. 11946. Springer.

Alamdari, P. M., Navimipour, N. J., Hosseinzadeh, M.,

Safaei, A. A., and Darwesh, A. (2017). A sys-

tematic study on the recommender systems in the e-

commerce. In IEEE Access. IEEE.

Ammar, W. B. H., Chaabouni, M., and Ghezala, H. B.

(2020). Recommender system for quality educational

resources. In Kumar V., Troussas C. (eds) Intelligent

Tutoring Systems. Springer.

Burke, R. (2002). Hybrid Recommender Systems: Survey

and Experiments.

Cai, X., Hu, Z., Zhao, P., Zhang, W., and Chen, J. (2020). A

hybrid recommendation system with many-objective

evolutionary algorithm. In Expert Systems with Appli-

cations, Vol. 159. ScienceDirect.

Cano, E. and Morisio, M. (2017). Hybrid recommender

systems: A systematic literature review. In Intelligent

Data Analysis, Vol. 21, No. 6.

Hybrid Recommendation Systems: A State of Art

287

Cao, L., Ma, B., Zhou, Y., and Chen, B. (2018). Design

and implementation of writing recommendation sys-

tem based on hybrid recommendation. In IEEE Ac-

cess, Vol. 6. IEEE.

Chen, R., Hua, Q., Chang, Y., Wang, B., Zhang, L., and

Kong, X. (2017). A systematic study on the recom-

mender systems in the e- commerce. In IEEE Access,

Vol. 8. IEEE.

Chen, R., Hua, Q., Chang, Y., Wang, B., Zhang, L., and

Kong, X. (2018). A survey of collaborative ﬁltering-

based recommender systems: from traditional meth-

ods to hybrid methods based on social networks.

Dhawan, S. (2019). Comparision of recommendation sys-

tem approaches. In COMITCon’19. IEEE.

Do, H. Q., Le, T. H., and Yoon, B. (2020). Dynamic

weighted hybrid recommender systems. In ICACT’20.

IEEE.

Duzen, Z. and Aktas, M. (2016). An approach to hybrid per-

sonalized recommender systems. In INISTA’16. IEEE.

Ghazanfar, M. and Prugel-Bennett, A. (2010a). An im-

proved switching hybrid recommender system using

naive bayes classiﬁer and collaborative ﬁltering. In

IAENG’10.

Ghazanfar, M. A. and Prugel-Bennett, A. (2010b). A

scalable, accurate hybrid recommender system. In

KDDM’10. IEEE.

Gong, J., Ye, Y., and Stefanidis, K. (2020). A hybrid rec-

ommender system for steam games. In Flouris G. et

al. (eds) Information Search, Integration, and Person-

alization. Springer.

Gulzar, Z., Leema, A. A., and Deepak, G. (2018). Pcrs: Per-

sonalized course recommender system based on hy-

brid approach. In ICSCC’17. ScienceDirect.

Idrissi, N., Zellou, A., Hourrane, O., Bakkoury, Z., and

Benlahmar, E. H. (2019a). Addressing cold start chal-

lenges in recommender systems: Towards a new hy-

brid approach. In SmartNets’19. IEEE.

Idrissi, N., Zellou, A., Hourrane, O., Bakkoury, Z., and

Benlahmar, E. H. (2019b). A new hybrid-enhanced

recommender system for mitigating cold start issues.

In ICIME’19. ACM.

Kitchenham, B. and Charters, S. (2007). Guidelines for per-

forming systematic literature reviews in software en-

gineering. In EBSE Technical Report EBSE-2007-01.

Springer.

Kumar, P., Kumar, V., and Thakur, R. (2018). A new ap-

proach for rating prediction system using collabora-

tive ﬁltering. In Iran Journal of Computer Science,

Vol. 2, No. 2. Springer.

Li, C., Wangb, Z., Caoa, S., and He, L. (2018a). Wlrrs: A

new recommendation system based on weighted lin-

ear regression models. In Computers and Electrical

Engineering, Vol. 66. ScienceDirect.

Li, X., Xing, J., Wang, H., Zheng, L., Jia, S., and Wang, Q.

(2018b). A hybrid recommendation method based on

feature for ofﬂine book personalization. In Journal of

Computers.

Maihami, V., Zandi, D., and Naderi, K. (2019). Proposing

a novel method for improving the performance of col-

laborative ﬁltering systems regarding the priority of

similar users. In Physica A: Statistical Mechanics and

its Applications. ScienceDirect.

Mansur, F., Patel, V., and Patel, M. (2017). A review on

recommender systems. In ICIIECS’17. IEEE.

Najmani, K., habib, B. E., Sael, N., and Zellou, A. (2019).

A comparative study on recommender systems ap-

proaches. In BDIoT’19. ACM.

Nikzad–Khasmakhi, N., Balafar, M., and Feizi–Derakhshi,

M. R. (2019). The state-of-the-art in expert recom-

mendation systems. In Engineering Applications of

Artiﬁcial Intelligence, Vol. 82.

Pandey, A. K. and Rajpoot, D. S. (2016). Resolving cold

start problem in recommendation system using demo-

graphic approach. In ICSC’16. IEEE.

Patel, B., Desai, P., and Panchal, U. (2017). Methods of

recommender system: a review. In ICIIECS’17. IEEE.

Prakash, K., Asad, F., and Urolagin, S. (2019). User and

item preference learning for hybrid recommendation

systems.

Santos, N. M. . H. D. K. V. . O. C. (2014). Recommender

Systems for Technology Enhanced Learning, Research

Trends and Applications. springer.

Sattar, A., Ghazanfar, M. A., and Iqbal, M. (2017). Build-

ing accurate and practical recommender system algo-

rithms using machine learning classiﬁer and collabo-

rative ﬁltering. In Computer Engineering and Com-

puter Science.

Shah, K., Salunke, A., Dongare, S., and Antala, K. (2017).

Recommender systems: An overview of different ap-

proaches to recommendations. In ICIIECS’17. IEEE.

Song, Y., Liu, S., and Ji, W. (2016). Research on person-

alized hybrid recommendation system. In CITS’17.

IEEE.

Tang, Q. and Wang, H. (2016). Privacy-preserving hybrid

recommender system.

Tian, Y., Zheng, B., Wang, Y., Zhang, Y., and Wu, Q.

(2019). College library personalized recommendation

system based on hybrid recommendation algorithm.

In Procedia CIRP, Vol. 83.

Tsolakidis, A., Triperina, E., Sgouropoulou, C., and Chris-

tidis, N. (2016). Research publication recommenda-

tion system based on a hybrid approach. In PCI’16.

ACM.

Wairegi, S., Mwangi, W., and Rimiru, R. (2020). A frame-

work for items recommendation system using hybrid

approach. In IST-Africa Conference. IEEE.

Wang, N., Zhao, H., Zhu, X., and Li, N. (2019). The review

of recommendation system.

Zhang, S., Yao, L., Sun, A., and Tay, Y. (2018). Deep learn-

ing based recommender system: A survey and new

perspectives. In IEEE Access. ACM.

ENASE 2021 - 16th International Conference on Evaluation of Novel Approaches to Software Engineering

288