Identification of Opinion Leaders and Followers in Social Media

Chun-Che Huang

, Li-Ching Lien

, Po-An Chen

, Tzu-Liang Tseng

and Shian-Hua Lin

Department of Information Management, National Chi Nan University, Puli, Taiwan, R.O.C.

Department of Industrial, Manufacturing and Systems Engineering, The University of Texas at El Paso,

El Paso, TX, U.S.A.

Department of Computer Science and Information Engineering, National Chi Nan University, Puli, Taiwan, R.O.C.

ywords: Opinion Leaders and Followers Identification, Matrix Analysis, Social Media, Web Analytics.

Abstract: In recent years, with the development of Web2.0, opinion leaders on the Web go up onto the stage and lead

the will of the people. Many time, government, private companies and even traditional news media need to

understand the opinion leaders’ ideas on the Web. Identifying opinion leaders and followers becomes a very

important study. To study the characteristics of opinion leaders and the impact of opinion leaders on followers,

our research evaluates whether every speaker in social media satisfies characteristics of opinion leader. The

characteristics of opinion leader and relationship between opinion leader and follower are studied. By

observing relational matrix, the interacting relations between users in social media are analysed and opinion

leaders and followers are identified.

1 INTRODUCTION

Nowadays the opinion leaders have significant

impact on the Internet. To recognize the opinion

leaders’ ideas is crucial from government, private

companies to traditional news media. Spanish scholar

Ramonet (2003) proposed that there should be the

fifth right, a public power to balance the increasingly

chaotic fourth right. In ‘Towards a Civil Society’

seminar in 2008, when discuss the role of blog in

improving social development and discussion

promotion, Blog is defined as the fifth right.

New network media such as Facebook, Twitter

and Taiwan local PTT as the fifth right have huge

users in discussing different issues. In these forums,

there are many opinion leaders with professional

knowledge to provide opinions to discuss. In this

way, there are many followers who could be

influenced by opinion leaders. However, the studies

related to opinion leader identification emphasize on

business sales rather than opinion leaders of social

policies. Previous studies utilize quantitative methods

to analyse and without considering from the

qualitative perspectives. It is difficult to accurately

identify the potential opinion leaders and followers

without exploring the relationship between opinion

leaders and followers.

This study surveys literature in Section II. Section

III proposes a methodology to identify the opinion

leaders and followers. Based on opinion leaders’

social network characteristics and whether their

speeches could get resonation, the matrix approach is

proposed to model the relationship between users and

identify opinion leaders set Section IV concludes this

study. In addition, this study contributes offering

three patterns to help the enterprises/government to

implement their corresponding strategy. For example

of general pattern, business take advantages of

opinion leaders to change follower behaviour;

therefore, their products/services can be promoted

effectively. According to Discussion pattern, the

opinion leaders and followers is not obvious such that

business can ignore them to decrease investment. For

example of blindly pattern, business can focus on

negotiation professional to improve recruiting the

opinion leaders and followers.

2 LITERATURE REVIEW

The word ‘opinion leader’ comes from study of

president election in 1940 by Lazarsfeld, Berelson,

and Gaudet (1948). They observed the behaviors of

voters in election and found that the election

advertisements were not as effective as opinion

180

Huang, C-C., Lien, L-C., Chen, P-A., Tseng, T-L. and Lin, S-H.

Identiﬁcation of Opinion Leaders and Followers in Social Media.

DOI: 10.5220/0006416801800185

In Proceedings of the 6th International Conference on Data Science, Technology and Applications (DATA 2017), pages 180-185

ISBN: 978-989-758-255-4

leaders. However, opinion leaders are defined

broadly. Opinion leaders are very important in social

network because they are capable to informally

influence others’ attitudes and behaviors (Cho,

Hwang & Lee, 2012, Li & Du, 2011; Li, Xing, Wang,

& Ren, 2012,; Venkatraman, 1989). Opinion leaders

usually have access to far more information on certain

topic and have profession in this topic. These people

are usually core of the social network and easy to get

along with. The roles of opinion leaders vary by

different themes, fields, cultures, social environments

and eras (Weimann & Brosius, 1994). Those who

receive the information and change their behaviors

and attitudes are called followers (Burt, 1999).

Compared with opinion leaders, the followers are not

as important. In other words, the followers are those

who will be influenced by opinion leaders and change

their attitudes and behaviors. (Li et al., 2012)

According to the study listed above, we can

conclude the properties of opinion leaders in the

following:

1. The life experience is rich and the

understanding of the knowledge is though.

Many of them are highly educated.

2. Strong social skills, strong connection with

broad masses, good reputation due to

professional and knowledgeable. They have

great influence and appealing power.

3. Sensitivity to the information, willingness to

accept new things, with innovative spirit.

In this study, we define the opinion leaders are those

who have a lot of followers. They are celebrities or

important leaders in some areas. They have good

reputation and higher education level. So their

speeches and behaviors could influence others. And

the followers are defined as those who are usually

influenced by opinion leaders’ speeches and

behaviors. As a result, the followers will change their

attitudes on the issues.

In previous studies, Ma and Liu (2014) used

SuperedgeRank algorithm to analyze 3 seed network

property. This was applied on identifying opinion

leaders in Fukushima nuclear issue. Cho et al. (2012)

analyze the role of opinion leaders in marketing

strategy with innovatively spreading social network

research. Leal, Hor-Meyll, and de Paula Pessoa

(2014) use interviews to study how opinion leaders

influence users’ purchase decision in virtual

communities. Li and Du (2014) propose a framework

by building key words base and apply this framework

to identify opinion leaders in Twitter users. William,

McMurray, Kurz and Lambert (2015) use key words

labels and social network analysis to analyze the

behaviors and attitudes of Twitter users towards

climate change. In summary, most previous studies

use methods such as social measure and self-report

which are based on questionnaire or interview to

identify opinion leaders. Few of them are based on

articles published by opinion leaders or social pattern

of opinion leaders and followers. Our study proposes

a method to analyze the articles published by the

users, users’ characteristics and their interactive

relationship. Combined with relational matrix

method, we identify opinion leaders and followers as

well as social pattern when they interact with each

other.

3 METHODOLOGY

In this study, the users’ published contents is

classified according to expert’s judgement (Table 1).

Positive speech means that the spokesman’s

statement is the same with responder’s attitude, the

opinions are the same. Meaningless speech means the

statement and spokesman are unrelated or

undetermined. Negative speech means spokesman’s

statement are different from responder’s attitude, the

opinions are different.

Table 1: Speech Contents Classification.

Contents Classification

Speech Contents Score

Uniform Speech

Meaningless Speech +0

Not uniform Speech -1

Next, the relationship between users is

represented by a matix formation. Matrix M is a

relational matrix between users. The row of matrix is

defined as , the column of the matrix is defined as j.

The count of users are set as C={1 … n}. The

elements in matrix T are count of responses and being

responded. The elements in matrix I are influence

factor between users and social community support

level.

M=[T|I]

Matrix T represents count of responses and being

responded between users.



∗

represents count of

responses between each pair of users.

n: count of users.





: count of user’s speech.

i: index of the user who responds to other people’s

opinion.

j: index of the user whose opinion is responded.

Matrix I represents users’ power to influence and

being influenced and social community support level.





: social community support level of the user.

Identiﬁcation of Opinion Leaders and Followers in Social Media

181

i: index of the user who influences to other people.

j: index of the user who is influenced.

In matrix T, ≠j, then 



means counts of

responses of i to j. When i=j, relationship 



means

counts of speech of i. In matrix I, when characteristic

and property ≠j, 



means influence of  on j. In

this study, the influence of  and j can be classified

into 3 different layers according to job title,

professional knowledge and social community

support level. Each criteria have three levels, no

influence (NI), general influence (GI), high influence

(HI).

Table 2: Influence Power Comparison Table.

Knowledge

status

Social Status

Social

Community

Support

low General high

Wrong

NI NI NI

Low

Social

Community

Support

Popular

NI NI GI

Professional

NI GI GI

Knowledge

status

Social Status

Social

Community

Support

low General high

Wrong

NI NI GI

Neutral

Social

Community

Support

Popular

NI GI GI

Professional

GI GI HI

Knowledge

status

Social Status

Social

Community

Support

low General high

Wrong

NI GI HI

High

Social

Community

Support

Popular

GI HI HI

Professional

HI HI HI

In this study, we compare and analyze two users

based on these three criteria. In table 2 influence

comparison table, we compare influence relationship

between users. For the higher job title, professional

knowledge or social community support level, the

influence power is high influence (HI). For the

general job title, professional knowledge and social

community support level, the influence power is

general influence (GI). For those with no job,

inaccurate professional knowledge or fair social

community support level, the influence power is no

influence (NI). Note that when compare these users,

we could only analyze the users’ speech contents and

professional knowledge. So in our study, we need not

to complete data for each user. We can list the part we

could not get or judge as missing. The missing part is

listed as blank.

In the Influence Power Comparison Table, the

Social Community Support refers to the power of the

forum and how close between the user and the media.

For example, if users can easily get and pass the

message to the media, we define the user has high

social community. Besides, if a particular user posts

the article that are read by large of users, we also

defined it is high social community.

In matrix I, when characteristic and property are

equal i=j, then the relationship I(ij) means the social

community support level of this user. This study

classifies social community support into four levels.

For the well-known, well followed social media, it is

classified as social community support high (H)

(

Table 3). For the less-known, less followed social

media, it is classified as low (L). However, due to

some users’ privacy setting or anonymous discussion,

our study could not know this user’s social

community support level. Our study defines these

users as social community support level missing (O).

After defining influence power, compared with

expert justified speech counts in table 1 classification

of speech contents, our study lists the user with

negative speech count as non –follower. At the same

time, if the contents being responded is negative, it is

listed as Non-opinion leader.

Table 3: Social Community Support Level.

In this study, there are three ways to show the

social relationship between opinion leaders and

followers in relational matrix, as shown in below

social community pattern.

Social community Pattern:

General Pattern:

In this pattern, the opinion leaders usually have

high social community support and their posts

are very professional. So they can broadly

influence many followers. This is most common

pattern in general social community.

Discussion Pattern:

The appearance of discussion pattern comes

from discussion space provided for social

community platform users. Users may propose

different specific advice when discuss with each

other and thus influence other users.

Social Community Support Level

Category Code

Low Social

Community Support

Neutral Social

Community Support

High Social Community

Support

Missing O

DATA 2017 - 6th International Conference on Data Science, Technology and Applications

182

Blindly Follow:

This pattern comes into exist because the

followers follow the opinion leaders by their

personal charm. Generally, followers in this

pattern do not care about the correctness of the

contents published by opinion leaders.

This study defines the opinion leader and follower

identification problem as searching of two mutual

influenced relational matrices, where the speech

counts and influence power in two matrices are

separated following the limitations:

(1) The users with negative count of responses

are not followers.

(2) Consecutive speech is viewed as one speech.

(3) Users’ speech contents should be first

justified by experts.

(4) The influence power of users’ speech is

decided by speech contents and social

community support level.

(5) The high social community support between

users is judged as “GI” or “HI”

(6) The low social community support is judged

as “NI”

Next, the solution approach to resolved the problem

is proposed:

Figure 1: Matrix modeling.

Step 1 Crawling data: According to forum (fan pages,

posts) in social media according to a particular topic,

data are crawled including text and responses

information.

Step 2 Convert data to the matrix formulation:

Crawled data are cleaned and converted to matrix T

from social media according to counts of interaction

between users (Figure 1).

Step 3 Remove meaningless users: Remove

meaningless users from matrix. The users with no

speech and never being responded are removed

(Figure 2).

Figure 2: Remove meaningless users.

Step 4 Label users’ class: Label users’ social

community support level based on collected materials

and put into matrix T according to table 3 social

community support level (Figure 3).

Figure 3: Label users’ class.

Step 5 Influence power analysis: According to

influence power level in table 2 influence power

comparison, analysis the users’ influence power

based on experts’ judgement and input the influence

power into matrix I (Figure 4).

Step 6 Sorting: Based on counts of speeches,

responses, being responded and influence power,

move forward the related users. The users with

greater counts of responses and being responded are

listed in the more front.

Step 7 Grouping: Base on Clustering Identification

(CI) Algorithm, the Group are formed. Group users in

matrix T and I by the users’ relationship. When

influence power between users is high or there are

responses between users, we classify these users in

Identiﬁcation of Opinion Leaders and Followers in Social Media

183

the same group. If there is no relationship, we do not

group them.

Figure 4: Influence power analysis.

Step 8 Identification: In each group, based on opinion

leaders’ definition in this study, define interaction

between users, opinion leaders and followers.

At final, the ABC network platform is used as an

example to explain application of our study method

in practical. The platform is open since 02/10/2015.

This example is based on national energy conference

held by bureau of energy, ministry of economic affair.

After four months’ discussion, this national energy

conference produces a great of common opinions and

other opinions. This study utilizes relational matrix to

analyze green energy related issues in ABC network

platform. The users in ABC network platform could

use anonymous style to discuss or use Facebook,

Google or Yahoo and so on social media username

and password to log in. This study collects, classifies

and analyzes the discussion data in ‘Stable Energy’

issue on ABC network platform.

After remove meaningless users, M=[T|I] (partial

data in Figure 5&6) is formulated. Note that the

related users is moved forward based on count of

users’ speech, responses, being responded and

influence power. The users with greater count of

responses and being responded are listed more front.

In Step 7, Group A {1, 2, 3, 4, 5, 6, 18, 20, 21, 24},

Group B {2, 16, 24, 26, 27, 34, 36}, Group C {17, 18,

19}, Group D {19, 20, 21, 36}, Group E {2, 27, 28,

29, 30, 36}, Group F {7, 20, 26}, Group G {2, 9, 27,

31, 34}, Group H {7, 27}, Group I {7, 8, 15}, Group

J {21, 22, 23}, Group K {25, 26}, Group G {34, 35}

are formed. In Step 8, for example of Group A, both

response and responded contents of user 1 have quite

influence power.}

So user 1 is an opinion leader. User

18, 24, 21, 3, 5 and 6 are positive towards user 1’s

opinion and they are followers. There are discussions

in this group, so this is a discussion pattern (Figure 6).

Figure 5: Matrix elements conversion.

Figure 6: Matrix elements conversion.

Figure 7: Matrix elements conversion.

Note that opinion leader is labeled with . Follower

is labeled with . Mutual influenced discussion

pattern is labeled with .

In group D, the influence power of user 19’s

responses to user 20 and 21 are high. But the

responses influence is not high. User 19 and 20 have

discussion relationship, so this is a discussion pattern

(Figure 8).

DATA 2017 - 6th International Conference on Data Science, Technology and Applications

184

Figure 8: Group D social pattern analysis.

Note that refers Mutual influenced discussion

pattern is labeled with .

All other groups are analyzed. User 1 and 36 are

summarized as opinion leaders. Group A, C, D, F and

H are discussion pattern.

Table 4: Analysis Results.

Opinion Leader Discussion Pattern Group

User 1, 36 A, C, D, F, H

The results are validated by domain experts and

shows that this study can effectively identify opinion

leaders and define social community pattern in the

social communities, where users’ support level could

not be obtained because users disagree with each

other, users are high controversial, less persons are

involved in discussion or many users are anonymous.

Through observation of social community pattern

among users, we could know users will not support a

user’s opinions because the user has many speeches.

4 CONCLUSIONS

This study utilizes relational matrix to find the

relationship between opinion leaders and followers.

Create criteria of social community support level and

influence power level between users. Combined with

experts’ judge to identify opinion leaders and

followers. According to social community support

level and influence power level in this study, utilize

relational matrix to identify opinion leaders in green

power issue in the social communities. Utilize our

study method to identify social pattern between users.

Then we can know, when identify opinion leaders, the

social community support level and posted contents

are very important to identify opinion leaders. Users

could have positive and negative opinions toward the

issue. Only considering connection between users’

posts are not enough. Even the users’ post a lot, if they

can’t get support from others, they could not be

defined as opinion leaders.

Future study and suggestions:

Currently, this study is applied on green energy low

carbon issue. In future, this could be applied in

marketing filed. Opinion leaders and followers could

get comments and reviews of products from social

community.

This study proposes social community support level

and influence power level. Also we apply relational

matrix to analyze relationship between users and

social community pattern. If this could be applied in

the social media with many information and highly

discussed. This study could be more complete.

This study analyzes with static information. In

future, we could collect dynamics information to

analyze and get instant identification.

REFERENCES

Aguado, E., & Burt, J. E., 2007. Understanding weather

and climate, Pearson Prentice Hall Upper Saddle River,

NJ, USA.

Burt, R. S. 1999., The social capital of opinion leaders. The

Annals of the American Academy of Political and

Social Science, 566(1), 37-54.

Cho, Y., Hwang, J., & Lee, D., 2012. Identification of

effective opinion leaders in the diffusion of

technological innovation: A social network approach.

Technological Forecasting and Social Change, 79(1),

97-106

Katz, E., & Lazarsfeld, P. F. 1955., Personal Influence: The

part played by people in the flow of mass

communications ,Transaction Publishers

Lazarsfeld, P. F., Berelson, B., & Gaudet, H., 1948., The

people's choice. New York: Columbia UniversityPress,

1948

Leal, G. P. A., Hor-Meyll, L. F., & de Paula Pessôa, L. A.

G., 2014., Influence of virtual communities in

purchasing decisions: The participants' perspective.

Journal of Business Research, 67(5), 882-890.

Li, F., & Du, T. C., 2011, Who is talking? An ontology-

based opinion leader identification framework for

word-of-mouth marketing in online social blogs.

Decision Support Systems

Li, F., & Du, T. C., 2014., Listen to me—Evaluating the

influence of micro-blogs. Decision Support Systems.

Ma, N., & Liu, Y., 2014. , SuperedgeRank algorithm and

its application in identifying opinion leader of online

public opinion supernetwork. Springer

Weimann, G., & Brosius, H.-B., 1994. Is there a two-step

flow of agenda-setting? International Journal of Public

Opinion Research

Williams, H. T., McMurray, J. R., Kurz, T., & Lambert, F.

H., 2015. Network analysis reveals open forums and

echo chambers in social media discussions of climate

change. Global Environmental Change

Identiﬁcation of Opinion Leaders and Followers in Social Media

185