Identiﬁcation of Social Inﬂuence on Social Networks and Its Use in

Recommender Systems: A Systematic Review

Lesly Alejandra Gonzalez Camacho

and Solange Nice Alves-Souza

Departamento de Engenharia de Computac¸

ao e Sistemas Digitais,

Escola Polit

ecnica da Universidade de S

ao Paulo, S

ao Paulo, Brazil

Keywords:

Social Network, Social Inﬂuence, Recommender System, Systematic Literature Review.

Abstract:

Currently the popularization of social networks has encouraged people to have more interactions on the internet

through information sharing or posting activities. Different social media are a source of information that can

provide valuable insight into user feedbacks, interaction history and social relationships. With this information

it is possible to discover relationships of trust between people that can inﬂuence their potential behavior when

purchasing a product or service. Social networks have shown to play an important role in e-commerce for

the diffusion or acquisition of products. Knowing how to mine information from social networks to discover

patterns of social inﬂuence can be very useful for e-commerce platforms, or for streaming of music, tv or

movies. Discovering inﬂuence patterns can make item recommendations more accurate, especially when there

is no knowledge about a user’s tastes. This paper presents a systematic literature review that shows the main

works that use social networking data to identify the most inﬂuential set of users within a social network and

how this information is used in recommender systems. The results of this work show the main techniques used

to calculate social inﬂuence, as well as identify which data are the most used to determine inﬂuence and which

evaluation metrics are used to validate each of the proposals. From 80 papers analyzed, 14 were classiﬁed as

completely relevant regarding the research questions deﬁned in the SLR.

1 INTRODUCTION

The mass adoption of network communication tech-

nologies has signiﬁcantly expanded the population

that are aware of social networking concepts and in-

terested in the data produced there. A number of peo-

ple currently actively manage an explicit network of

virtual friends, contacts, associates and internet ad-

dresses that make up their family, social, and pro-

fessional lives. An easy and common way to see

how highly connected people exchange information

by email messages sent from person to person. The

notion of “friends of friends” is easily illustrated in

the features of social media applications such as Face-

book, Instagram, Twitter, LinkedIn, which offer ser-

vices explicitly called “social networking”. The suc-

cessions of information shared within these networks

illustrate the modern way in which interaction be-

tween people has shifted to computer-mediated chan-

nels of communication. Social networks are ser-

vices that allow people to browse and to connect with

https://orcid.org/0000-0002-9387-8351

https://orcid.org/0000-0002-6112-3536

friends of their friends (Hansen et al., 2011).

Social media encourages users to have more in-

teractions on the Internet by sharing information or

posting activities. On social networks, the relation-

ship can be deﬁned by the number of social informa-

tion shared (eg post, comments, likes) between users

(Hendry et al., 2017). Social networks make more

visible the ties and connections that have always con-

nected people, such as relationships between teams,

partnerships, tribes groups, alliances, companies, in-

stitutions, organizations, among others; types of rela-

tionships that before the existence of social networks

were less apparent (Hansen et al., 2011).

With the information available in social networks,

it is possible to detect knowledge that may be useful

to offer better products and services to users in differ-

ent ﬁelds. For example, in the ﬁeld of Recommender

Systems (RS), it can provide more accurate item rec-

ommendations, in part thanks to social networks, be-

cause the mass information produced by online users

in social networks creates new opportunities to help

researchers and developers better understand the user

preferences (Li and Xiong, 2017). The Netﬂix en-

234

Camacho, L. and Alves-Souza, S.

Identiﬁcation of Social Inﬂuence on Social Networks and Its Use in Recommender Systems: A Systematic Review.

DOI: 10.5220/0009829702340241

In Proceedings of the 9th International Conference on Data Science, Technology and Applications (DATA 2020), pages 234-241

ISBN: 978-989-758-440-4

terprise uses data from social network, such as Face-

book, to discover the friend circle using preferences

(ex. titles watched), or crossing people data from dif-

ferent countries with similar proﬁles to recommend

foreign titles.

The large volume of data generated daily makes it

harder for users to ﬁnd what interests them, because

of the multiple items/product options available on the

internet. Due to this, RS plays a very important role

in suggesting items that allow users to have a positive

experience when making a purchase or purchasing a

product (Lalwani et al., 2015; Deng et al., 2014).

RSs are software tools and techniques that aim to

provide users with suggestions for items that are ef-

fectively useful to their needs. RS is referred to as

“item” which is recommended to an user. An RS

usually focuses on generating recommendations for

an item type (eg, music, movies, electronics, news),

but it can also be for a set of types, or miscellaneous

products, such as e-commerce (Prando et al., 2017).

Recommendations are all customized to provide use-

ful and efﬁcient product or product suggestions to the

user (Ricci et al., 2011).

RS methods that use social information assume

that social network data can discover ties of friend-

ship, trust, or inﬂuence among users. From these ties

of trust, or inﬂuence, users are more likely to develop

a greater afﬁnity for items purchased by their social

ties. Different approaches have been proposed (Yang

et al., 2018; Wu et al., 2018; Li and Xiong, 2017;

Zhang et al., 2017; Zhou, J et al., 2017) to obtain the

degree of social inﬂuence that some people cause on

others based on information from social networks.

This paper presents a Systematic Literature Re-

view (SLR) approaching different proposals from re-

cent research found in the literature analyzing social

network information to calculate social inﬂuence and

how this information can be useful in RS (Wang et al.,

2016).

This work is divided as follows; Section 2 presents

the main aspects of previous works in the literature

review regarding correlated themes. Section 3 details

the process followed to carry out the SLR here pre-

sented. Section 4 shows the SLR results. Section 5

introduces the main aspects of the papers classiﬁed as

relevant to the SLR and the conclusion of the work is

shown in section 6.

2 RELATED WORKS AND

BACKGROUND

Different approaches have been used by e-commerce,

entertainment, services or content platforms to rec-

ommend products to their users (Thilagam, 2016).

Among the main approaches used in RS, the literature

highlighted Content-Based Filtering (CB), Collabora-

tive Filtering (CF) and Hybrid Filtering.

Content Based Recommender System (CB-RS)

generates recommendations based solely on the user’s

proﬁle, that is, on the set of favorite items, or that were

searched in the past by the user (Huang et al., 2016).

Therefore, the system learns to recommend items that

are similar to those one liked in the past.

Collaborative Filtering Recommender System

(CF-RS) is characterized by recommending to a user

items that other users with similar preferences have

enjoyed in the past. Similarity between users is calcu-

lated based on the behavior of the ratings they made

on the items. These behaviors allow predicting fu-

ture assessment behaviors for other items. CF-RS is

considered the most popular and widely implemented

technique in RS (Ricci et al., 2011; Desrosiers and

Karypis, 2011).

Hybrid systems are based on combining the tech-

niques, thus using the advantages of one that can over-

come the deﬁciencies of the other and improve the re-

sults of the recommendation (Ricci et al., 2011).

An emerging topic in the literature is the social

recommender systems, which is based on the as-

sumption that popular items adopted by the user’s

trusted friends can be recommended to him/her (Li

and Xiong, 2017). Some researches in the litera-

ture (Xiushan and Dongfeng, 2017; Lian et al., 2016;

Wang et al., 2016) have shown that using social net-

work information can be a good resource for discover-

ing relationships of trust, or inﬂuence, which can help

mitigate popular problems that happen regardless of

the approach taken in RS, such as Cold-start and Data

Sparsity, and improve item recommendations in RS.

Wu et al. (2013) carried out a review related with

social media applications, associated to where infor-

mation is created and how it is exchanged in applica-

tions such as e-commerce, content-sharing sites, so-

cial network sites, virtual community, and collabora-

tive projects. In relation to social inﬂuence, the au-

thors presented some examples of studies that aim

to identify inﬂuential users using some information

from social networks and techniques such as Page-

Rank. They also commented about some heuristics

employed to calculate the social inﬂuence and even

about graph models. However, its approach is a lit-

tle generic, it was not shown, for example, the type

of graph technique that were used to model social in-

ﬂuence, nor showed techniques that can be applied in

RS, nor addressed the type of information employed

by previous researchers to calculate the social inﬂu-

ence. In social networks there are different data and

Identiﬁcation of Social Inﬂuence on Social Networks and Its Use in Recommender Systems: A Systematic Review

235

each of them present particular features that can be

more useful to calculate social inﬂuence. Then, to

know about the data type employed in a particular re-

search should be important to propose new ways of

calculating social inﬂuence.

Li et al. (2018) showed proposals of techniques

to calculate social inﬂuence. Despite of the authors

have related several techniques, they did not concen-

trated in proposals that consider inﬂuence inside of

a group of friends, besides they did not consider ap-

plications as RS. Considering that closest friend usu-

ally have more afﬁnity, the inﬂuence among them is

stronger (Gonzalez-Camacho, L.A. and Alves-Souza,

S. N., 2018) and more effective mainly for RS. The

SLR presented here, focuses on researching of tech-

niques to calculate inﬂuence in a group of friends in

social networks, identifying the most inﬂuential indi-

vidual into the group. It is also showed the kind of

data would be interesting in the calculation of social

inﬂuence to improve the recommendations.

3 PROCEDURE FOR

SYSTEMATIC LITERATURE

REVIEW

For a more objective literature review, SLR was pro-

posed following the guidelines of Kitchenham and

Charters (2007), who say that an SLR is a method

that is previously deﬁned and followed to identify, se-

lect, evaluate and synthesize works related to the re-

search theme. This SLR has been divided into three

steps: Planning, Execution, and Summary. Each of

these steps contains a set of steps that were followed.

3.1 Research Questions and Search

Strings

The main research question we tried to answer was:

• There are techniques to identify in social networks

the most inﬂuential user and/or set of users; was

this information used to improve the recommen-

dation?

As search string was deﬁned:

• (”recommender system”) AND (”social net-

work”) AND (”user social relation” OR ”group

users” OR ”user inﬂuence”)

The databases chosen to search the papers were:

• Scopus, IEEE, ACM e Web of Science.

The choice of these bases was based on the trust of

their content by the computing area and the access to

the full text of the papers, which is guaranteed by the

University.

3.2 Inclusion and Exclusion Criteria

Table 1 shows the inclusion and exclusion criteria de-

ﬁned for the initial selection of papers.

Table 1: Inclusion and exclusion criteria.

Inclusion Exclusion

The research uses so-

cial networking infor-

mation to determine the

most inﬂuential friend

The research is not

written in English

The research is related

to recommender sys-

tems

The research identiﬁes

the type of user social

relationship within the

social networks

The research was

published in a journal

or conference between

2017 and 2019

3.3 SLR Process

Figure 1: SLR process diagram.

Figure 1 presents a diagram summarizing the process

followed in the method employed to perform the SLR.

Each part of these process is detailed below:

• Planning: At this stage, a protocol was elabo-

rated, with the research question and related key-

words. The search strings were elaborated and the

DATA 2020 - 9th International Conference on Data Science, Technology and Applications

236

search databases were chosen. Additionally, the

inclusion and exclusion criteria were determined

for the research selection. Finally, ﬁve evaluation

questions were deﬁned to assist in the ﬁnal ex-

traction of the most relevant research and they are

speciﬁed in Section 4.

• Execution: This phase is related to retrieving pa-

pers that satisfy the search conditions. For this,

the search strings were applied to the selected

search bases, along with the criteria deﬁned in the

planning stage. For selecting the most relevant

works in relation to the research question, the re-

search was divided into two further steps:

– Selection: only the title and abstract of the pa-

pers were read. Following the inclusion and

exclusion criteria, the papers were selected for

full reading or discarding. Filtering to deter-

mine the relevance of the paper was also used.

– Extraction: The papers selected in the selection

stage were completely read and evaluated to es-

timate their importance within the scope of the

research.

• Summary: At this stage, we obtained the ﬁnal re-

sults of the SLR. As a result, the main character-

istics of these works are summarized.

4 SLR RESULTS

Figure 2 shows the results from applying the search

strings to the databases already mentioned. This Fig-

ure shows the number of papers retrieved by publica-

tion type. In total, 80 papers were found, of which 47

were published in conferences, 32 in journals and 1

in a book. We preferred not to summarize by search

base, because it was not the objective of this work to

evaluate or to compare these databases.

Figure 2: Number of papers by type of publication.

Table 2 shows the number of ﬁltered papers for

each step of the execution phase, ending with the total

number of papers classiﬁed as the most relevant.

Table 2: Number of papers found and selected in the execu-

tion phase.

Resume Number of papers

Total found 80

Duplicates 2

Pre-selected 32

Selected for full reading 24

Final extraction 14

In the extraction phase to evaluate and classify pa-

pers in terms of their importance to SLR, ﬁve ques-

tions were elaborated based on the main research

question as follows:

• Q1- Does the research identify communities to de-

ﬁne inﬂuence?

• Q2- In the research is the community/group iden-

tiﬁed by any similarity calculations among users?

• Q3- Does the research identify the most inﬂuential

individual within the group / social network?

• Q4- Is this inﬂuence used to improve the recom-

mendation of any item?

• Q5- Is this inﬂuence directly related to the degree

of friendship?

The score for each question was determined as a

binary value (0-1), 1 was assigned if the paper meets

the Qi assessment question, or 0 if the paper does not

(Gonzalez-Camacho, L.A. and Alves-Souza, S. N.,

2018). This score allowed classifying the papers se-

lected for reading in order of relevance and ﬁnally,

selecting those that could effectively answer the main

research question. Each paper was evaluated against

each of the 5 questions (Q1, Q2, ..., Q5) listed above.

The sum of the scores (S) determined the ﬁnal grade

and the classiﬁcation of the paper, i.e., (S = 5): com-

pletely relevant, (3 <= S <= 4): partially relevant

and (S <= 2): not relevant. Table 3 shows the ones

classiﬁed as completely or partially relevant.

Table 3 also shows which of the questions the pa-

per answered assertively. Papers are ordered by the

score (highest to lowest), and the year of publication.

As a result, 14 papers were considered to be able to

answer the main research question.

5 KEY ASPECTS OF MAIN

PAPERS

Tables 4 and 5 summarize the techniques and infor-

mation set used in the 14 papers selected at the end of

this SLR.

Table 4 shows, for each paper, which techniques

were used to determine the social inﬂuence on the so-

cial network. From the set of selected papers it was

Identiﬁcation of Social Inﬂuence on Social Networks and Its Use in Recommender Systems: A Systematic Review

237

Table 3: Classiﬁcation of papers by evaluation question.

References Q1 Q2 Q3 Q4 Q5 S

Diaz-Agudo et al.

(2018)

3 3 3 3 3 5

Yang et al. (2018) 3 3 3 3 3 5

Liu et al. (2018) 3 3 3 3 4

Hendry et al.

(2017)

3 3 3 3 4

Li and Xiong

(2017)

3 3 3 3 4

Bhowmick et al.

(2018)

3 3 3 3

Ma et al. (2018) 3 3 3 3

Wu et al. (2018) 3 3 3 3

Jianqiang et al.

(2017)

3 3 3 3

Xing et al. (2017) 3 3 3 3

Sumith et al.

(2017)

3 3 3 3

Zhang et al. (2017) 3 3 3 3

Zhou, J et al.

(2017)

3 3 3 3

Zhou et al. (2017) 3 3 3 3

noticed that, in general, the techniques could be clas-

siﬁed into two approaches: Graphs and Heuristics /

Graphs. Graphs include the use of centrality measures

to determine social inﬂuence, while heuristics/graphs

involve the use of centrality measures associated with

heuristics proposed by the authors, or other different

techniques from those commonly used. The graph ap-

proach was divided into centrality measures: Eigen-

vector (Eigen), Degree (Deg), Closeness (Close), Be-

tweenness (Betw). While the Heuristics / Graphs ap-

proach was divided into Graph-Based (Gp-B), Page

Rank Based (Pr-B) and Other.

From the selected papers, 3 performed graph mod-

eling to analyze the behavior of social network in-

formation and applied different graph-centric met-

rics to determine social inﬂuence. The other 10 pa-

pers proposed heuristic algorithms for calculating in-

ﬂuence. Some of these were based on graph tech-

niques to model social interaction, but included other

ways to assess inﬂuence, corresponding to Table 4,

the technique classiﬁed as ”graph-based” (Gp-B). Pa-

pers which applied different techniques from those

commonly used and proposed new ways to identify

inﬂuence were classiﬁed as “others”.

When a social network is modeled by graphs, each

person is modeled as a vertex and their connections

to other vertices are referred to as edges. For ana-

lyzing a network, it is necessary to deﬁne some met-

rics that allow, for example, comparison with other

networks, tracking changes over time, or determining

the relative position of individuals and groups within

it (Hansen et al., 2011). Centrality metrics are ways

to analyze of social networks by graphs, which allow

capturing the importance of a vertex (node) within the

network, based on some criteria. These metrics allow

identifying people who are most important (have the

most connections) by the position in which they are

allocated. For example, some people are allocated at

the edge, or periphery of the network, while others are

allocated more to the center and connected to all other

people (Hansen et al., 2011).

The following centrality metrics provide quantiﬁ-

able measures (Hansen et al., 2011):

• Degree centrality (Deg): is characterized by the

number of connections linked to a vertex. When

the network is directed, this measure is divided

into two: In-degree, which is the number of con-

nections that point into a vertex. Out-degree is

the number of connections that originate from one

vertex and point to other vertices. Degree central-

ity is generally similar to a measure of popularity,

but inefﬁcient as it cannot differentiate between

quantity and quality. For example, by this mea-

sure, it is not possible to differentiate between a

relationship with the president of the republic and

a relationship with the state university.

• Betweenness centrality (Betw): This is a measure

of how often a given vertex is on the shortest path

between two other vertices. It is a measure that

allows evaluating how much removing a person

would break connections between others in the

network. This gives the highest score to those that

serve as a ”bridge” to connect to other people in

the network. For example, measuring the shortest

distance between people who are not neighbors,

but are neighbors to other neighbors, and so on.

• Closeness centrality (Close): . With this measure,

it is assumed that vertices can only transmit mes-

sages or inﬂuence their existing connections. A

low value means the extent a person is directly

connected, or ”jump away” from most of the oth-

ers in the network. For example, vertices in very

peripheral locations may have high closeness cen-

trality scores, which point to the number of hops,

or connections, they need to make to connect to

others far away in the network.

• Eigenvector centrality (Eigen): Allows connec-

tions to have a variable value; thus, connecting to

some vertices has more beneﬁts than connecting

to others. The Eigenvector view is more sophisti-

cated than the other measures of centrality, a per-

son with few connections could score very high if

those connections were very well connected, e.g.,

those that have a high number of messages ex-

DATA 2020 - 9th International Conference on Data Science, Technology and Applications

238

Table 4: Techniques used to calculate social inﬂuence.

Reference

Graphs Heuristics / Graphs

Eigen Deg Close Betw Gp-B Pr-B Other

Bhowmick et al. (2018) 3

Diaz-Agudo et al. (2018) 3

Liu et al. (2018) 3

Ma et al. (2018) 3 3 3 3

Wu et al. (2018) 3

Yang et al. (2018) 3

Hendry et al. (2017) 3

Jianqiang et al. (2017) 3

Li and Xiong (2017) 3

Xing et al. (2017) 3

Sumith et al. (2017) 3

Zhang et al. (2017) 3

Zhou, J et al. (2017) 3

Zhou et al. (2017) 3 3 3

Total de artigos 2 2 2 2 4 2 5

Table 5: Social data types used to calculate social inﬂuence.

Reference

Social data

Fw Fwe Ps R-Ps Lk Cm Mt Other

Bhowmick et al. (2018) 3 3

Diaz-Agudo et al. (2018) 3 3 3 3

Liu et al. (2018) 3 3

Ma et al. (2018) 3 3 3

Wu et al. (2018) 3 3

Yang et al. (2018) 3 3 3

Hendry et al. (2017) 3 3 3 3

Jianqiang et al. (2017) 3 3 3

Li and Xiong (2017) 3 3 3 3

Xing et al. (2017) 3 3 3 3 3 3

Sumith et al. (2017) 3 3 3

Zhang et al. (2017) 3 3 3 3

Zhou, J et al. (2017) 3 3 3 3 3 3 3 3

Zhou et al. (2017) 3 3 3

Total Papers 6 6 5 9 6 9 3 7

changed.

• Page rank (Pr-B): such as Eigenvector, page rank

measures connections based on their qualiﬁcation.

It is an algorithm used by the Google search en-

gine for information retrieval.

Table 5 highlights the seven data types, whose def-

inition is given below, used by each paper to deter-

mined social inﬂuence. The data type ”other” refers to

ones that appeared less frequently, such as: time when

information is propagated in a social network, statis-

tics on the relevance of published content, or type of

social interaction not speciﬁed. The Results in Table

5 indicate which data appear to have the most weight

when assessing social inﬂuence on social networks.

As can be veriﬁed, Re-posting, Commenting, and Fol-

lowee were the most used to estimate social inﬂuence.

• Follower (Fw): set of users who follow a partic-

ular user. For example, if user A follows user B

(A → B), A is part of B’s followers (Xing et al.,

2017). It can also be interpreted as a friendship

tie.

• Followee (Fwe): set of users that a particular user

follows. For the previous case, user B is part of

the individuals followed by A (Xing et al., 2017).

It can also be interpreted as a friendship tie.

• Posting (Ps): set of information shared by a user

in the social network.

• Re-posting (R-Ps): can be interpreted as informa-

tion posted by one user, which has already been

posted by another.

• Likes (Lk): posted information that is highlighted

Identiﬁcation of Social Inﬂuence on Social Networks and Its Use in Recommender Systems: A Systematic Review

239

by social network users, indicating that they liked

the published content.

• Comments (Cm): are the opinions made by users

of the social network to the published content.

• Mentions (Mt): This is when in a comment, a user

names another user, or particular users.

To evaluate the performance of proposed algo-

rithms to determine social inﬂuence of the individu-

als in a social network most papers used an empirical

assessment method (Liu et al., 2018; Ma et al., 2018;

Wu et al., 2018; Yang et al., 2018; Hendry et al., 2017;

Sumith et al., 2017; Zhang et al., 2017; Zhou et al.,

2017). However, there is no widely accepted mea-

sure in the literature, at least as far as this research

has reached, that serves to verify the performance of

such algorithms.

Yang et al. (2018); Li and Xiong (2017); Zhou

et al. (2017) used the social inﬂuence calculation to

make some recommendations. Jianqiang et al. (2017)

could adapt the measures Precision, Recall and F1

based on the identiﬁcation of a set of reference inﬂu-

ential individuals to prove the efﬁciency of its algo-

rithm. Zhou, J et al. (2017); Bhowmick et al. (2018);

Jianqiang et al. (2017) have even not implemented the

social inﬂuence to recommend items, they instead es-

timated the efﬁciency of social inﬂuence in the pro-

posed approach.

6 CONCLUSION

This work investigated how the information produced

from the different interactions between users in social

networks can be very useful in RS. Speciﬁcally, pa-

pers published in journals and conferences between

2017 and 2019 were collected; their main objec-

tive was to identify the most inﬂuential individual or

group of individuals from mining information from

social networks.

Different approaches have been proposed to calcu-

late social inﬂuence and the graph was the technique

mostly employed to model social interaction. Some

of these works used centrality measures to estimate

social inﬂuence, although many others proposed new

ways for this calculation.

The advantage of the graph technique to model re-

lationship among users in a social network is that it

allows to not only quantifying these links but also to

qualify them, mainly when heuristics and other tech-

niques are added. The social inﬂuence model can be

different depending on what it is employed for. For

example, in a recommender system, the social data

types used in the social inﬂuence model can receive

different weights depending on what is being recom-

mended.

SLR highlighted the most used social data types to

estimate social inﬂuence. The papers selected showed

that the most commonly used data were: number of

posts re-posted, number of comments and number

of follow-ups that users had in their social network.

These three types of data could have a greater weight

when proposing a model for calculating social inﬂu-

ence.

Finally, the social inﬂuence model proposed was

concluded to preferably evaluated empirically.

The result of this SLR can be used as a basis for

determining the most relevant information that can

be extracted from social networks to model different

forms of social inﬂuence that can be used in recom-

mender systems to improve the accuracy of recom-

mendations.

ACKNOWLEDGEMENTS

The authors are grateful for the support given by

ao Paulo Research Foundation (FAPESP). Grant

#2014/04851-8, and the support given by Ita

u Uni-

banco S.A. trough the Ita

u Scholarship Program, at

the Centro de Ci

encia de Dados (C

D), Universidade

de S

ao Paulo, Brazil.

REFERENCES

Bhowmick, A. K., Sai Bharath Chandra, G., Singh, Y., and

Mitra, B. (2018). Constructing Inﬂuence Trees from

Temporal Sequence of Retweets: An Analytical Ap-

proach. In 2018 IEEE International Conference on

Big Data (Big Data), pages 624–633.

Deng, S., Huang, L., and Xu, G. (2014). Social

network-based service recommendation with trust

enhancement. Expert Systems with Applications,

41(18):8075–8084.

Desrosiers, C. and Karypis, G. (2011). A Comprehen-

sive Survey of Neighborhood-based Recommendation

Methods, chapter 4. Springer US.

Diaz-Agudo, B., Jimenez-Diaz, G., and Recio-Garcia, J. A.

(2018). SocialFan: Integrating social networks into

recommender systems. Proceedings - International

Conference on Tools with Artiﬁcial Intelligence, IC-

TAI, 2018-November:171–176.

Gonzalez-Camacho, L.A. and Alves-Souza, S. N. (2018).

Social network data to alleviate cold-start in recom-

mender system: A systematic review. Information

Processing & Management, 54(4):529–544.

Hansen, D. L., Shneiderman, B., and Smith, M. A. (2011).

Visualizing and Interpreting Social Media Networks

with NodeXL, chapter 3, pages 39–41.

DATA 2020 - 9th International Conference on Data Science, Technology and Applications

240

Hendry, Su, Y.-J., and Chen, R.-C. (2017). A New Method

for Identifying Users Interest for Personalized Rec-

ommendations. In Tsai P.-W. Watada J., K. N., edi-

tor, Proceedings - 2016 3rd International Conference

on Computing Measurement Control and Sensor Net-

work, CMCSN 2016, pages 84–87. Institute of Electri-

cal and Electronics Engineers Inc.

Huang, T. C. K., Chen, Y. L., and Chen, M. C. (2016). A

novel recommendation model with Google similarity.

Decis. Support Syst., 89:17–27.

Jianqiang, Z., Xiaolin, G., and Feng, T. (2017). A New

Method of Identifying Inﬂuential Users in the Micro-

Blog Networks. IEEE Access, 5:3008–3015.

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

performing Systematic Literature reviews in Software

Engineering Version 2.3. Engineering, 45(4ve):1051.

Lalwani, D., Somayajulu, D. V. L. N., and Krishna, P. R.

(2015). A community driven social recommendation

system. Proc. - 2015 IEEE Int. Conf. Big Data, IEEE

Big Data 2015, pages 821–826.

Li, C. and Xiong, F. (2017). Social recommendation with

multiple inﬂuence from direct user interactions. IEEE

Access, 5:16288–16296.

Li, K., Zhang, L., and Huang, H. (2018). Social Inﬂuence

Analysis: Models, Methods, and Evaluation. Engi-

neering, 4(1):40–46.

Lian, D., Ge, Y., Zhang, F., Yuan, N. J., Xie, X., Zhou,

T., and Rui, Y. (2016). Content-aware collaborative

ﬁltering for location recommendation based on human

mobility data. Proc. - IEEE Int. Conf. Data Mining,

ICDM, 2016-Janua:261–270.

Liu, J., Hao, F., and Wang, Y. (2018). Discovering In-

ﬂuential Areas According to Check-In Records and

User Inﬂuence in Social Networks. In 2018 IEEE

20th International Conference on High Performance

Computing and Communications; IEEE 16th Inter-

national Conference on Smart City; IEEE 4th Inter-

national Conference on Data Science and Systems

(HPCC/SmartCity/DSS), pages 1151–1155.

Ma, Q., Luo, X., and Zhao, M. (2018). User Behaviour Net-

work Based User Role Mining of Web Event. In 2018

14th International Conference on Semantics, Knowl-

edge and Grids (SKG), pages 211–217.

Prando, A. V., Contratres, F. G., Souza, S. N. A., and De

Souza, L. S. (2017). Content-based Recommender

System using Social Networks for Cold-start Users. In

9th International Joint Conference on Knowledge Dis-

covery, Knowledge Engineering and Knowledge Man-

agement (KDIR 2017), pages 181–189.

Ricci, F., Rokach, L., and Shapira, B. (2011). Recommender

Systems Handbook. Springer US.

Sumith, N., Annappa, B., and Bhattacharya, S. (2017). A

holistic approach to inﬂuence maximization. Springer

International Publishing.

Thilagam, P. S. (2016). Alleviating Data Sparsity and

Cold Start in Recommender Systems using Social Be-

haviour. In 2016 5th Int. Conf. Recent Trends Inf.

Technol.

Wang, X., Lu, W., Ester, M., Wang, C., and Chen, C. (2016).

Social Recommendation with Strong and Weak Ties.

In Conf. Inf. Knowl. Manag., pages 5–14. ACM Press.

Wu, J., Sun, H., and Tan, Y. (2013). Social media research:

A review. Journal of Systems Science and Systems

Engineering, 22(3):257–282.

Wu, L., Yang, B., Jian, M., Sang, J., Zhang, D., and Zhang,

L. (2018). MPPR: Multi Perspective Page Rank for

User Inﬂuence Estimation. In 2018 IEEE Interna-

tional Conference on Big Data and Smart Computing

(BigComp), pages 25–29.

Xing, L., Ma, Q., and Jiang, L. (2017). Microblog user rec-

ommendation based on particle swarm optimization.

China Communications, 14(5):134–144.

Xiushan, X. and Dongfeng, Y. (2017). A novel matrix

factorization recommendation algorithm fusing social

Trust and Behaviors in micro-blogs. In 2017 IEEE

2nd Int. Conf. Cloud Comput. Big Data Anal., pages

283–287. IEEE.

Yang, S., Shi, L., and Liu, L. (2018). Community Detection

Method Based on User Inﬂuence Probability and Sim-

ilarity. In 2018 Sixth International Conference on Ad-

vanced Cloud and Big Data (CBD), pages 183–190.

Zhang, C., Lu, T., Chen, S., and Zhang, C. (2017). In-

tegrating Ego, Homophily, and Structural Factors

to Measure User Inﬂuence in Online Community.

IEEE Transactions on Professional Communication,

60(3):292–305.

Zhou, Y., Li, Y., Wang, Z., Luo, Y., and Yang, X. (2017).

Identiﬁcation of inﬂuential spreaders in geo-social

network. In 2017 25th International Conference on

Geoinformatics, pages 1–4.

Zhou, J, Wu, G., Tu, M., Wang, B., Zhang, Y., and Yan,

Y. (2017). Predicting user inﬂuence under the envi-

ronment of big data. In 2017 2nd IEEE International

Conference on Cloud Computing and Big Data Analy-

sis, ICCCBDA 2017, pages 133–138. Institute of Elec-

trical and Electronics Engineers Inc.

Identiﬁcation of Social Inﬂuence on Social Networks and Its Use in Recommender Systems: A Systematic Review

241