Exploring political discussions by Korean twitter users: A look at opinion leadership and homophily phenomenon

Exploring political discussions byKorean twitter users

A look at opinion leadership and homophilyphenomenonMyunggoon Choi

Department of Interaction Science, Sungkyunkwan University, Seoul,South Korea

Yoonmo SangDepartment of Radio-Television-Film, The University of Texas at Austin,

Austin, Texas, USA, andHan Woo Park

Department of Media & Communication, YeungNam University,Gyeongsan-si, South Korea

AbstractPurpose – The purpose of this paper is to provide a network analysis of Twitter discussions aboutMyung-Bak Lee, a former president of South Korea, to gain a better understanding of the dynamics ofthe public opinion exchange on Twitter.Design/methodology/approach – Opinion leaders in the discussion network were identified byconsidering the longitudinal distribution of tweets containing the former president’s name, and threetypes of messages (“followings,” “mentions,” and “retweets”) were analyzed using data collectedfrom November 1, 2011, to April 20, 2012. The sample included 26,150 Twitter users and 892,034relationships reflecting three types of messages.Findings – The results indicate that the discussion about President Myung-Bak Lee was dominatedby liberal Twitter users who already had considerable influence both online and offline. In addition,Twitter users were unlikely to interact with other users with opposing political views.Research limitations/implications – Almost all of the opinion leaders identified in the study heldliberal political views, and liberal Twitter users dominated the discussion network. In addition, theKorean Twitter network showed the presence of the homophily phenomenon, implying that opinionleaders’ influence within the Twitter network was limited to other users sharing the same political views.Further, political views of opinion leaders were skewed toward a particular political stance withoutnecessarily representing the opinion of the general public, possibly hindering the democratic process.Originality/value – This study tests the homophily thesis in the context of Twitter users in Koreaand contributes to the literature on Twitter-based political discourse by identifying opinion leaders inKorean Twitter networks and examining the phenomenon of homophily within those networks.Keywords Webometrics, Korea, Twitter, Communication, NodeXL, Opinion leaders, Homophily,Social network analysisPaper type Case study

Aslib Journal of InformationManagementVol. 66 No. 6, 2014pp. 582-602© Emerald Group Publishing Limited2050-3806DOI 10.1108/AJIM-11-2012-0089

Received 21 November 2012Revised 24 July 201328 October 201327 January 20148 April 201415 May 2014Accepted 20 May 2014

The current issue and full text archive of this journal is available atwww.emeraldinsight.com/2050-3806.htm

Many thanks go to research assistants, Ji-Young Park, Ji-Young Kim, Seoung-Cheol Choi,Ji-Yeun Kim, Yun-CheolHeo, and XanatMeza. Also, the authors are grateful to the Editor andanonymous reviewers for their comments on earlier versions. The corresponding authoracknowledges that this research is partially supported by the WCU (World Class UniversityWebometrics Institute) program of the National Research Foundation of Korea, funded by theMinistry of Education, Science and Technology (No. 515-82-06574). The first two authorscontributed equally to this paper.

582

AJIM66,6

Dow

nloa

ded

by U

nive

rsity

of

Tex

as L

ibra

ries

At 1

1:48

02

Nov

embe

r 20

14 (

PT)

IntroductionAfter first rising to prominence as a tool for political engagement during the 2010 localelections in South Korea (hereafter “Korea”), Twitter has become pivotal in shaping thecountry’s political landscape. According to a recent survey by GlobalWebIndex, aglobal market research firm, as of the third-quarter of 2013, 56 percent of Koreaninternet users had Twitter accounts, and 22 percent used Twitter in the past month[1].In the US, in the months leading up to the 2010 midterm elections, 22 percent of allinternet users were using Twitter or other social networking sites (SNSs) such asFacebook and MySpace to connect to campaigns or the election itself (Smith, 2011).

Recognizing the potential of SNSs as an efficient communication platform,individuals and government agencies have been making increased use of the tool(Chung et al., 2014; Hsu et al., 2013; Khan et al., 2014). Many studies have shown thatthe social network structure has considerable influence on the diffusion of newsand information, the formation of people’s attitudes and the behavior of communitymembers (Rogers, 2003; Valente, 2010; Rho, 2014). By allowing internet users todistribute information, share opinions, and manage social ties, SNSs have expandedto the arena of political discourse (Gueorguieva, 2007; Rainie and Smith, 2012).According to a recent study by the Pew Research Center, approximately 75 percentof SNS users had friends who posted at least some content representing their politicalview and 37 percent had occasionally posted political messages (Rainie and Smith,2012). Given the increasing popularity of SNSs, the public is more likely to haveknowledge about certain issues. In addition, online discussions offer users greaterexposure to a cross section of political viewpoints (e.g. Kim, 2011).

However, in terms of contributing to a pluralistic democracy, the role of SNSsremains poorly defined. Previous studies have suggested that online social networksmay be fragmenting into subgroups of like-minded individuals (An et al., 2013;Sunstein, 2007). This notion, built on supportive evidence from “selective exposure”thesis and “homophily” theory, underscores the potential for polarization of viewpoints(Chang and Park, 2012; Monge and Contractor, 2003; Park and Thelwall, 2008;Sunstein, 2007). Given that SNSs have emerged only in recent years, there is a need tobetter address the question of whether the homophily thesis is applicable to Twitternetworks.

This study examines the network structure of political discussions on Twitterto gain a better understanding of how Twitter networks influence the online flow ofnews and information. More specifically, the study identifies “opinion leaders,” thoseindividuals who are more “connected” than others and thus are more likely to influencethe flow of information (Monge and Contractor, 2003; Valente, 2010; Valente andPumpuang, 2007).

Finally, this study examines whether network participants can be divided intosubgroups, potentially accelerating the diffusion of information on the one handand possibly limiting information flow and diverse viewpoints on the other (Monge andContractor, 2003; Rogers, 2003; Valente, 2010). More specifically, the study considerspolitical discussions disseminated on Twitter about a former Korean president bydrawing on social network theory and its methods. Previous studies have examinedthe role of Twitter in political discourse mainly in Western contexts or in the context ofpolitical upheavals in the Middle East (e.g. Harrow and Harp, 2012; Howard andHussain, 2013; Small, 2011; Smith, 2011). In this regard, understanding the ways inwhich Koreans use Twitter for political discourse should provide valuable data for aglobal comparison.

583

By Koreantwitter users

Dow

nloa

ded

by U

nive

rsity

of

Tex

as L

ibra

ries

At 1

1:48

02

Nov

embe

r 20

14 (

PT)

Literature reviewUsing SNSs for political discourseSNS use has emerged as a prominent social trend in recent years. Previous researchhas suggested that SNSs may function as a platform for political engagement (e.g.Gueorguieva, 2007). Although a review of the literature on the relationship betweensocial media use and political engagement goes beyond the purpose of this study, abrief summary of the literature is provided to lay the groundwork for identifying theunderlying structure of political discourse on Twitter.

According to a 2012 survey, 66 percent of adult internet users in the USA were SNSusers (Rainie and Smith, 2012). A main goal of SNSs is the formation and maintenanceof online communities that consist of individuals who share similar interests andconcerns (Ellison et al., 2007). SNSs are web-based services that enable users to buildsocial capital within their networks (Ellison et al., 2007; Gruzd et al., 2011) and becomeconnected to other users based on shared interests, views, or activities. Such participatoryand interactive platforms are changing patterns of communication by facilitatingconnections between users, encouraging interactions and direct participation (Chou et al.,2009). The relationships between citizens, politics, and the media have changeddramatically through the rise of countless blogs, wikis, and other user-generated contentand collaborative platforms (Bruns, 2005).

SNSs have essentially become integral to the political process (Rainie and Smith,2012), including national campaigns (Denton and Kuypers, 2008). Many commentatorshave noted that SNSs can be an effective tool for motivating the public’s politicalengagement, particularly during campaigns (Gueorguieva, 2007; Owen, 2008; Smithand Rainie, 2008). However, Kushin and Yamamoto (2010) pointed out that previousstudies of the relationship between social media use and political engagement haveproduced mixed results. Some have claimed that social media have beneficial effects onpolitical engagement (Valenzuela et al., 2009), whereas others have found no significantrelationship between the two (Gil de Zúñiga et al., 2009; Zhang et al., 2010).

Opinion leadership on twitterIn the last several decades, researchers have explained the information-sharingbehavior of individuals based on Katz and Lazarsfeld’s (1955/2006) two-step flowtheory, which is also useful for highlighting the role of “opinion leaders” who facilitatethe dissemination of media messages to audiences. This theory posits that one’s trustedclose friends or relatives may have more influence on one’s opinion than a mediamessage (Katz and Lazarsfeld, 1955/2006).

Later, Katz (1957) proposed three criteria that distinguish “opinion leaders” fromnon-leaders, arguing that one’s influence is related first, “to the personification ofcertain values (who one is); second, to competence (what one knows); and third, tostrategic social location (whom one knows)” (p. 73). Given the importance of “opinionleaders” in disseminating ideas and information, the concept of opinion leadership hasbeen widely adopted by a broad range of disciplines such as political science,marketing, diffusion studies, and sociology. More recently, Rogers (2003) definedopinion leadership as having a disproportionate amount of power over others’ attitudesand behaviors.

Scholars have attempted to best define and measure opinion leadership (Weimannet al., 2007). Previous studies of opinion leadership have generally relied on subjectiveinformation derived from self-report surveys or interviews (Katz and Lazarsfeld, 1955/2006).

584

AJIM66,6

Dow

nloa

ded

by U

nive

rsity

of

Tex

as L

ibra

ries

At 1

1:48

02

Nov

embe

r 20

14 (

PT)

To address various limitations associated with subjectivity, the network analysis techniquehas been adopted to explore opinion leaders within the structure of social relationships(Xu et al., 2014). As part of such efforts, researchers have adopted sociometric[2] techniquesto identify opinion leaders (Valente and Pumpuang, 2007). For instance, previous studieshave examined the structure of connections between actors in a network and consideredopinion leaders to be those receiving a large number of sociometric choices (Valenteand Pumpuang, 2007). One of the most frequently used methods for identifying suchsociometric choices is by evaluating the network ties one receives from others (Valente,2010). In SNSs, network ties are often based on social relationships, the exchange ofresources, and the flow of information (Monge and Contractor, 2003; Valente, 2010).

Opinion leaders are more likely to take up strategically beneficial positions in anetwork (Monge and Contractor, 2003). The notion of centrality as a micro-levelindicator of one’s position in a given network is used to analyze opinion leadership(Monge and Contractor, 2003; Valente, 2010). Those individuals who show high levelsof centrality are more connected and thus occupy more central positions in the network(Monge and Contractor, 2003; Valente, 2010). Among various sociometric indicatorsof opinion leadership, indegree centrality, which is based on the number of ties onereceives from others, is employed most frequently (Valente, 2010).

Wu et al. (2011) tested two-step flow theory in the context of Twitter and foundthat almost half of all tweets originating from the media are disseminated through theso-called “opinion leaders,” that is, through network participants who are more widelylinked and exposed to media messages than their followers. In a similar vein, An et al.(2011) found that friends who follow particular media sources influence other friends’media exposure on Twitter.

As discussed earlier, some types of opinion leaders are readily recognizable basedon their popularity and prominent leadership position (Valente and Pumpuang, 2007).Therefore, it is reasonable to assume that traditional opinion leaders (i.e. offline leaders)influence their Twitter network. In this regard, the following research question isaddressed:

RQ1. Who are the opinion leaders on Twitter in the discussion on Korean PresidentMyung-Bak Lee and what factors characterize them?

Do Twitter users flock together?It is important to investigate whether Twitter networks expose members to differentpolitical views. If SNSs such as Twitter facilitate differential exposure, then there maybe important implications for the democratic process. Mutz (2002) pointed out that ifindividuals avoid exposing themselves to different political views and ignore anyinformation that challenges their preexisting opinion, then they are less likely to beopen to opposing viewpoints. This can be problematic in that a society may find itspolitical process in a state of gridlock.

The theory of selective exposure suggests that “individuals often systematicallyprefer information that is consistent with their beliefs, attitudes, or decisions and, incontrast, neglect inconsistent information” (Fischer et al., 2008, p. 382). Many studieshave examined individuals’ tendency to engage in selective exposure (Stroud, 2007,2010; Sunstein, 2007), and recent studies have increasingly focused on understandingthe causes and consequences of selective exposure (e.g. Stroud, 2007, 2010). Sunstein(2007) argued in his book Republic.com 2.0 that filtering technologies can facilitateselective exposure on the internet. Eventually, selective exposure is likely to result in

585

By Koreantwitter users

Dow

nloa

ded

by U

nive

rsity

of

Tex

as L

ibra

ries

At 1

1:48

02

Nov

embe

r 20

14 (

PT)

highly fragmented and polarized opinions. Stroud (2010) empirically demonstratedthat partisan selective exposure, together with partisanship, can further polarize theAmerican public. Mutz (2006) noted that “selective exposure is likely to account forthe fact that the networks of highly partisan and politically active people are moredominated by like-minded discussants” (p. 44). Taken together, previous researchhas suggested that selective exposure can hinder the formation of well-informed andhigh-quality public opinion.

The theory of homophily proposes that individuals are more likely to interactand associate with like-minded ones (Monge and Contractor, 2003), particularly in thecontext of online networks. Here filtering technologies can help like-minded individualsidentify and interact with one another (Sunstein, 2007). Chang and Park (2012)suggested that heterogeneous groups within online social networks tend to befragmented and that the fragmentation of these networks may lead to the polarizationof opinions.

In the specific context of Twitter, whether and to what extent the homophily thesisis applicable remains an open question because of the relative newness of Twitter.According to a recent study by the Pew Research Center (2011), political discourse onTwitter is more likely to be opinionated and often more negative concerning politicalcandidates than that on blogs or in the mainstream press. Kim (2011) examinedthe ways in which individuals’ use of SNSs influences their exposure to differentpolitical views, verifying that SNS use increases the likelihood of being exposed to“cross-cutting” political views.

However, a considerable number of studies have also reported that cross-ideologicalinteractions among SNSs users may be limited or impeded by the homophilyphenomenon. Conover et al. (2011) examined more than 250,000 tweets generated bymorethan 45,000 users during the six weeks prior to the 2010 US congressional midtermelections and found that retweet network to be highly polarized based on Twitter users’political alignment. However, they found no such phenomenon in the mention network.Gruzd (2012) also found the phenomenon of political polarization on Twitter inconjunction with weak evidence of cross-ideological connections during the 2011Canadian Federal Election. Himelboim et al. (2013) verified that, at least with respectto several controversial political topics, the homophily phenomenon occurs on Twitter.Based on their findings, Himelboim et al. (2013) concluded that meaningfulcross-ideological interactions are likely to be limited by the homophily phenomenon.Although these studies have provided a better understanding of the characteristics ofTwitter networks, further research is needed to better comprehend whether Twitterfacilitates communication between individuals with different political views or reinforceshomophily. In addition, few studies have extended the issue of homophily on Twitter toAsian contexts (e.g. Hsu and Park, 2012; Kim and Park, 2012). This gap in the literatureraises the question of whether the homophily thesis is applicable to the Twitter context.In this regard, the following research question is addressed:

RQ2. Can the Twitter-based network be divided into subgroups with similarpolitical views and how are the subgroups connected?

MethodNetwork theory and twitter networksThe network analysis technique offers a useful framework for understanding thedynamics of public opinion in social networks. It is well documented that the mass

586

AJIM66,6

Dow

nloa

ded

by U

nive

rsity

of

Tex

as L

ibra

ries

At 1

1:48

02

Nov

embe

r 20

14 (

PT)

media such as newspapers and television can play an important role in framing publicdebates and influencing the formation and evolution of public opinion (e.g. Chong andDruckman, 2007). However, previous studies of public opinion have paid relatively littleattention to the role of individuals in shaping public discourse and policies. It is clearthat identifying and capturing interpersonal communication that occurs in social mediaenvironments provide a valuable opportunity to better understand the dynamicsof public opinion for controversial social issues. As a result, an increasing number ofstudies have made use of the network analysis technique.

The network analysis technique involves a set of data collection and analysisprocedures designed to examine social relationships between various types of socialactors, including individuals, groups, organizations, and countries (Valente, 2010). Thesocial network analysis (SNA) enables researchers to examine social actors’ influenceon a network and relationships between actors by quantifying and visualizing socialties and the overall structure of formal and informal networks.

This study focuses on Twitter, a widely used microblogging service that allowsits users to post and read short messages (Golbeck et al., 2010; Takhteyev et al., 2012).Bruns and Burgess (2011) noted that Twitter performs two functions: social networkingthrough “friending” or “following” and the large-scale sharing and diffusion ofinformation. The convergence of social networking and information streams enablesTwitter to differentiate itself from other SNSs. This convergence is “what underpinsTwitter’s unique significance for journalism; and any serious evaluation of user activitieson Twitter must reflect on the structural aspects of this convergence” (Bruns andBurgess, 2011, p. 3). Twitter is unique in that its main goal ignores users’ preexisting ties(Takhteyev et al., 2012). Twitter users can post messages (tweets) and forward orcomment on other users’ messages (retweets). In recent years, Twitter users have usedhashtags, which consist of brief keywords or abbreviations with a prefixed hash symbolfor effective communication with an ad hoc community sharing the same concerns orinterests regarding a particular topic (Bruns and Burgess, 2011).

Data collectionTo identify users within a Korean Twitter network showing opinion leadership traits,those who listed the full name of Myung-Bak Lee, a former president of Korea, wereconsidered. For data collection purposes, only those tweets written in Korean wereincluded because the purpose of this study is to identify opinion leaders within KoreanTwitter networks and examine the phenomenon of homophily in the Korean context.

Tweets including the Korean word “ ,” the full name of the former president,were collected using NodeXL, an open-source network analytic tool that can collect andvisualize network data from various sources such as e-mail messages, Flickr, YouTube,and Twitter, from November 1, 2011, to April 20, 2012[3]. NodeXL offers two ways toimport data from social media and allows the researcher to import data directly fromsocial media or to use the NodeXL Network Server, which enables the downloading ofTwitter networks in an automatic and periodic manner. In this study, the NodeXLNetwork Server was employed. To collect data through the NodeXL Network Server, itis necessary to have an XML configuration file with several options such as searchterms, different types of relationships, and the maximum number of users for eachquery. In this study, three types of relationships were included, and the limit set byTwitter on the maximum number of users for each query (1,000 users per each request)was observed.

587

By Koreantwitter users

Dow

nloa

ded

by U

nive

rsity

of

Tex

as L

ibra

ries

At 1

1:48

02

Nov

embe

r 20

14 (

PT)

The application for data collection automatically retrieved tweets and userinformation from Twitter for those users mentioning “Myung-Bak Lee” and others whofollowed them. Twitter restricts access to a maximum of 1,500 tweets per query and 350requests per hour under Twitter Search API 1.0 (in chronological order). Because ofseveral technical limitations in collecting data from the Twitter Search API, we couldnot retrieve Twitter data when the waiting time lasted more than an hour or when morethan 350 calls were requested. We collected data automatically at 09:00 a.m., and as aresult, we obtained network data (consisting of 26,150 Twitter users and 892,034relationships in the discussion on the Korean president) for 49 days. The relationshipsconsisted of three types: 3,093 mentions, 62,495 retweets, and 826,446 followings.These numbers were reasonable and manageable for a snapshot of the network.

User data were analyzed by NodeXL, and their “following,” “retweet,” and “mention”relationships were considered in analyzing the data. Given that the followingrelationship is more ritual than the mention relationship, which requires morecommitment (Yoon and Park, 2014), previous studies have argued that the former maynot accurately explain the user’s influence (Cha et al., 2010; Leavitt et al., 2009). In thisregard, many researchers have highlighted differences between various types ofrelationships (see Conover et al., 2011). However, previous studies have shown thatcultural differences may have considerable influence on SNS users’ communicationstyles and relationships (Cho and Park, 2013; Khan et al., 2014). Given that reciprocity ishighly respected in relationships in Korean culture, reciprocal respect may also play animportant role in Twitter networks (see Cho and Park, 2012). Lee et al. (2011) found thatinfluential Twitter users in Korea are likely to mention other Twitter users’ names backif their names are mentioned because of the unwritten rule of social etiquette. That is,if someone is following another on Twitter, following that user back is consideredpolite, even though there is no written rule. Gruzd et al. (2011) found a moderatecorrelation between mutual and interaction networks and concluded that “people whoare mentioned in the same tweets are also likely to be [an influential user’s] mutualfollowers” (p. 1308). Based on these findings, the proposed research questions wereaddressed by focusing not only on following/follower relationships but also on retweet/mention relationships.

MeasurementOpinion leadership. In this study, indegree centrality, which is based on the number ofties received from other actors and widely used as an indicator of opinion leadership,was used (Valente, 2010). The higher the indegree centrality of an actor, the morecentral the actor’s position in the network is, indicating greater trust, authority, andpower (Monge and Contractor, 2003; Valente, 2010). Here NodeXL was used to calculatethe indegree centrality of all users in the network. The Results section discusses howthe list of opinion leaders was obtained based on their indegree centrality.

Network density. Few studies have provided an empirical analysis of the effectsof the internet or SNSs such as Twitter on political discourse in the context of Korea.That is, little is known about whether Twitter fosters participatory democracy orfragmentation/polarization. To address this issue, network ties between liberal Twitterusers, between conservative Twitter users, and between liberal and conservativeTwitter users were analyzed. More specifically, network density was evaluated by thefrequency of communication between liberal and conservative Twitter users throughmentions, followings, and retweets.

588

AJIM66,6

Dow

nloa

ded

by U

nive

rsity

of

Tex

as L

ibra

ries

At 1

1:48

02

Nov

embe

r 20

14 (

PT)

Gephi, an open-source network software program (available at http://gephi.org/),was used to examine and visualize subgroups. The analysis was limited to those daysshowing above-average network density and thus focused on 20 of the 49 availabledays. If network density is too low, then it is difficult to examine the characteristics ofthe network. In this study, if the modularity score was 0, then no interaction betweenusers was assumed. In addition, those clusters deviating from the center of the networkand accounting for o1 percent of the whole network were excluded from the analysis.Network density indicates the degree of network cohesion (Haythornthwaite, 1996)[4]and is the number of relationships in the network presented as a proportion of all possiblerelationships (Wasserman and Faust, 1994). Valente (2010) noted that “[d]ensity is alsooften calculated on subgroups within the network” (p. 130). High network density meansthat many possible relationships are actually present in the network. In this study,for example, there were heavy interconnections between Twitter users through eitherfollowing/follower or mention/retweet relationships. UCINET 6 was used to calculate andcompare network density within and between clusters[5]. Table I shows the stepssystematically followed in this study.

ResultsOpinion leaders in the context of twitterTo identify opinion leaders in each network, the indegree centrality of all users in thenetwork was measured, and the one with the highest indegree centrality was extractedeach day for which data were collected. Demographic characteristics of influentialTwitter users were obtained from their Twitter profiles. In addition, opinion leaders’occupations and political views were examined based on their Twitter profiles and theNaver People Search, a service that provides personal information on celebrities inpolitics, entertainment, and sports (Table II).

Within the discussion network, there were 25 users who were positioned at the topof the network in terms of indegree centrality, with the exception of “beast6_inf7”whose Twitter account no longer existed. These users included laypeople (ten),politicians (three), journalists (three), media outlets (three), novelists (two), professors(two), researchers (one), and critics (one).

For a more in-depth analysis of these actors, those showing the highest indegreecentrality in the discussion network at least two times were examined. The resultsindicate that, among the 25 users, nine appeared more than twice. All these usersexcept for “@coreacom,” who was unknown to the public, were popular individuals ororganizations in Korea. Noteworthy is that all were liberal in terms of their politicalstance.

As shown in Table II, two users appeared six times. One was a well-known liberaljournalist (@du0280), and the other was a high-profile researcher (@kennedian3).A novelist (@congjee), a liberal newspaper (@kyunghyang), and a liberal journalist(@welovehani) each appeared four times. A high-profile Twitter user (@coreacom), aliberal journalist (@dogsul), a liberal politician (@jk_space), and a liberal online newssite (@newsvop) each appeared two times. The results indicate that existing influentialactors (e.g. journalists and media outlets) were very likely to extend their offlineinfluence to online discussions.

Previous studies have documented that Koreans are likely to perceive user opinionsexpressed on Twitter as leaning toward a more liberal view in the matter of politicalissues (Kang, 2013; Lee and Kim, 2014). As shown in Table II, in terms of

589

By Koreantwitter users

Dow

nloa

ded

by U

nive

rsity

of

Tex

as L

ibra

ries

At 1

1:48

02

Nov

embe

r 20

14 (

PT)

No

Step

Descriptio

n

1Datacollection

NodeX

Lwith

TwitterSearch

API

1.0was

used

toextracttweets(singleton

messages,retw

eets,and

mentio

ns)including

thename“M

yung

-Bak

Lee”

andlists

offollowersof

thoseuserswho

createdthesetw

eets

from

Novem

ber1,2011

toApril20,2012

2Examiningdescriptive

characteristics

ofthediscussion

netw

ork

Networkfeatures

(indegree

centrality,

outdegreecentrality,

anddensity

)for

alla

vailabledateswerecalculated

byNodeX

L.Fo

rtechnicalreasons,itwas

possibleto

collect

data

only

for49

days

3Identifying

opinionleaders

ATwitter

user

who

over

theanalysisperiod

hadat

leasttwiceoccupied

thetoppositio

ninterm

sof

indegree

centrality

was

identifiedas

anopinionleader

inthediscussion

netw

ork.

There

werealtogether

nine

opinionleadersdu

ring

the

analysisperiod,and

theirpoliticalview

swereexam

ined

bytheirTwitter

profilesandthePeopleSearch

serviceoffered

by“N

aver,”therepresentativ

eportal

servicein

Korea

4Dissectingthenetw

ork

Allnetw

orks

foreach

dayweredividedinto

groups

basedon

themodularity

algorithm

throug

hGephi

(https://geph

i.org/).The

analysiswas

limitedto

thosedays

show

ingabove-averagenetw

orkdensity

5Determiningthepolitical

inclinationof

each

cluster

Aftersubclustersofdenselyinterconnected

nodeswereidentifiedthroug

hamodularity

analysis,10percentofallTwitter

userscomposing

each

subcluster

wereidentifiedbasedon

theirindegree

centrality.These

Twitterusers’politicalview

swerethen

codedby

twoindepend

entcoders.K

ripp

endorff’s

α(Kripp

endorff,2004)w

asused

toassess

inter-coder

reliability(0.8820).P

oliticalv

iewsof

theirneighb

orswereinferred

from

thoseof

centraln

odes

closeto

thoseneighb

ors.

Thisprocedurewas

basedon

egocentricsnow

ballsampling,atechniqu

einwhich

nodesareselected

bytheirproxim

itytoacentralnode.Thenclusters

werecollapsed

intoonelib

eralclusterandoneconservativ

eclustertocomparenetw

ork

density

betw

eenlib

eral

clusters,b

etweenconservativ

eclusters,and

betw

eenlib

eral

andconservativ

eclusters

6Co

rrelationof

wordfrequency

betw

eencore

andperiph

eral

groups

Toovercomethepotentialw

eakn

essof

inferringperiph

eral

nodes’political

view

sthroug

hegocentricsampling,

asemantic

analysiswas

complem

entarilycond

ucted.Wordfrequencywas

exam

ined

with

respecttotw

eetsgeneratedby

core

nodes(10%

)with

thehigh

estindegree

centrality,andthen

allw

ords

werecomparedto

thoseused

byperiph

eral

nodes(90%

)interm

sof

wordfrequency.

Thisprocedurewas

appliedto

both

liberal

andconservativ

egroups.H

ere

Geul-Jap-I(www.sejong.or.kr),a

Koreanlang

uage

softwarepackagethatallowsresearcherstogeneratewordfrequency

lists

andconcordance,was

employed.T

hencorrelations

betw

eenwords

used

bycore

nodes(10%

)with

thehigh

est

indegree

centralityandthoseused

byperiph

eral

nodes(90%

)wereevaluated

7Interactions

betw

eenlib

eral

andconservativ

eclusters

Toexam

inetheinteractionbetw

eenlib

eral

andconservativ

egroups,the

density

betw

eenandwith

inclusters

was

calculated

usingUCINET

Table I.Systematic steps

590

AJIM66,6

Dow

nloa

ded

by U

nive

rsity

of

Tex

as L

ibra

ries

At 1

1:48

02

Nov

embe

r 20

14 (

PT)

No.

Twitter

IDThe

numberof

times

theuser

hadthehigh

est

indegree

scorein

thediscussion

netw

ork

Followings

Followers

Tweets

Occup

ation

Political

view

1du

0280

6(12.24%)

17,391

17,596

17,484

Journalist

Liberal

2kenn

edian3

6(12.24%)

6,005

89,696

16,269

Researcher

Liberal

3cong

jee

4(8.16%

)381

267,531

15,012

Novelist

Liberal

4ky

ungh

yang

4(8.16%

)45,256

48,586

9,859

Media

outlet

Liberal

5welovehani

4(8.16%

)5,048

54,821

27,188

Journalist

Liberal

6coreacom

2(4.08%

)220,879

202,123

51,645

Unk

nown

Not

clear

7dogsul

2(4.08%

)68,725

139,033

65,106

Journalist

Liberal

8jk_space

2(4.08%

)24,830

28,476

11,370

Politician

Liberal

9newsvop

2(4.08%

)25,156

43,290

15,776

Media

outlet

Liberal

Table II.Opinion leaders

discussing Myung-BakLee from November 1,2011 to April 20, 2012

591

By Koreantwitter users

Dow

nloa

ded

by U

nive

rsity

of

Tex

as L

ibra

ries

At 1

1:48

02

Nov

embe

r 20

14 (

PT)

political views, among the nine major actors, five individuals and three organizationsheld liberal views. The political view held by “@Coreacom,” a high-profile Twitter user,was not clear, although this user likely held a liberal view. These results indicate that,in terms of political issues, liberal Twitter users had considerable influence on theTwitter network. These results provide support for the findings of previous studies(e.g. Hsu and Park, 2011; Hsu and Park, 2012). Hsu and Park (2011, 2012) examinedTwitter users in Korea and found that liberal messages are more likely to be widelycirculated and frequently referred to.

Mapping interactions within and between subgroupsSubclusters of densely interconnected nodes were identified through a modularityanalysis, and then 10 percent of all Twitter users in each subcluster were identifiedbased on their indegree centrality. Newman (2006) demonstrated that the modularityanalysis method is an effective technique for detecting and characterizing structures ofsubcommunities. First, political views of users with the highest indegree centralitywithin each subcluster were examined. For the determination of political views ofTwitter users with the highest indegree centrality, their Twitter profiles and actualtweets were evaluated because these profiles and tweets revealed the users’ politicalviews, given that they were actively participating in political discussions. TheseTwitter users’ political views were coded by two independent coders. Krippendorff’s α(Krippendorff, 2004) was used to assess inter-coder reliability (0.8820). Political views oftheir neighbors were inferred from those of central nodes close to those neighbors. Thisprocedure was based on egocentric snowball sampling, a technique in which nodes areselected by their closeness to a central node. Previous studies (e.g. Conover et al., 2011;Golbeck and Hansen, 2011; King et al., 2011) have shown that it may be possible to inferTwitter users’ political views by examining “the content, structure of their messages,and the social networks in which they are embedded” (Barberá, 2012, p. 6).

To overcome the potential weakness of inferring peripheral nodes’ political viewsthrough egocentric sampling, a semantic analysis was complementarily conducted.Word frequency was examined with respect to tweets generated by core nodes, andthen all words were compared to those used by peripheral nodes in terms of wordfrequency. This procedure was applied to both liberal and conservative groups. HereGeul-Jap-I, a Korean language software package that allows the researcher to generateword frequency lists and concordance, was employed. Then Pearson correlationsbetween words used by core nodes with the highest indegree centrality and those usedby peripheral nodes were evaluated. The results indicate high correlations betweencore and peripheral nodes at the 0.001 level (Table III). The correlations betweenwords used by liberal clusters and those used by conservative clusters were examined,and the results show low to moderate correlations at the 0.001 level.

After various subgroups were identified according to their political views andcollapsed into one conservative group and one liberal one, the questions of how and towhat extent the two groups interacted with each other were addressed. In so doing, theapplicability of the homophily thesis to the Twitter discussion on the Korean presidentwas examined. As discussed earlier, many studies have claimed that individuals aremore likely to share their opinions with like-minded others than to expose themselvesto opposing points of view.

In general, for the analysis period, liberal Twitter users (M¼ 76.34, SD¼ 10.88) wereclearly more active in discussing Myung-Bak Lee than conservative users (M¼ 18.01,

592

AJIM66,6

Dow

nloa

ded

by U

nive

rsity

of

Tex

as L

ibra

ries

At 1

1:48

02

Nov

embe

r 20

14 (

PT)

Num

berof

usersin

liberal

clusters

aNum

berof

usersin

conservativ

eclusters

aLiberalclusters

Conservativ

eclusters

Date(m

onth/day/

year)

Core

nodes

(10%

)Periph

eral

nodes

(90%

)Co

renodes

(10%

)Periph

eral

nodes

(90%

)Co

re(10%

)and

periph

eral

(90%

)nodes

Core

(10%

)and

periph

eral

(90%

)nodes

11/26/11

88789

11102

0.874***

0.824***

11/30/11

77692

28253

0.927***

0.773***

12/14/11

65583

39351

0.881***

0.848***

12/18/11

63565

24214

0.909***

0.679***

12/19/11

77695

23211

0.499***

0.844***

12/24/11

86771

330

0.871***

0.522***

12/25/11

89804

12112

0.912***

0.866***

12/26/11

81731

20184

0.866***

0.675***

01/01/12

85767

15136

0.899***

0.903***

01/14/12

91818

654

0.857***

0.836***

01/15/12

77689

871

0.786***

0.747***

01/21/12

80722

17151

0.624***

0.786***

01/22/12

79707

22202

0.823***

0.502***

01/27/12

84755

27246

0.851***

0.842***

01/29/12

73655

22198

0.879***

0.872***

02/09/12

64576

39348

0.909***

0.759***

03/10/12

92825

12105

0.581***

0.701***

03/18/12

84760

12108

0.882***

0.827***

03/24/12

97877

762

0.913***

0.840***

03/30/12

65587

32287

0.930***

0.934***

Notes

:a A

fter

theidentificationof

subclustersthroug

hamodularity

analysis,those

subclustersbelow

averagedensity

wereexclud

edfrom

thesemantic

analysis.***po

0.001.

Table III.Numbers of twitter users

in liberal andconservative clusters and

Pearson correlationcoefficients

593

By Koreantwitter users

Dow

nloa

ded

by U

nive

rsity

of

Tex

as L

ibra

ries

At 1

1:48

02

Nov

embe

r 20

14 (

PT)

SD¼ 9.48). Figure 1 visualizes changes in each network. White and gray nodesrepresent liberal users, whereas black nodes, conservative ones. As shown in Table IVand Figure 1, liberal clusters consistently occupied the most central positions inthe network.

As noted earlier, Twitter users were divided into liberal and conservative clusters.Steps to collapse clusters into one liberal cluster and one conservative one werefollowed to compare network density between liberal clusters, between conservativeclusters, and between liberal and conservative clusters. Here the question of whetherusers leaned toward one political direction (conservative or liberal) or interactedwith other users through mentions, retweets, or followings was addressed. In doing so,network density between liberal clusters, between conservative clusters, and betweenliberal and conservative clusters was examined (Figure 2).

An ANOVA was conducted to determine any significant differences in networkdensity between liberal clusters, between conservative clusters, and between liberaland conservative clusters. The results indicate a significant difference between thethree groups (F(2, 57) ¼ 24.8327, po0.0001). Conservative clusters (M¼ 0.0508,SD¼ 0.0329) were more densely connected than liberal ones (M¼ 0.0360, SD¼ 0.0065).In addition, the degree of connectedness between liberal and conservative clusters wasrelatively low (M¼ 0.0080, SD¼ 0.0026). This implies that Twitter users were morelikely to form relationships with politically like-minded users through mentions,followings, and retweets than with users with opposing political views. These resultsprovide support for the argument that individuals are more likely to interact with

2011-11-26 2011-11-30 2011-12-14 2011-12-18 2011-12-19

2011-12-24

2012-01-15 2012-01-292012-01-272012-01-222012-01-21

2011-12-25 2011-12-26 2012-01-01 2012-01-14

2012-02-09 2012-03-10 2012-03-18 2012-03-24 2012-03-30

Figure 1.Temporal changesin the network structureover time

594

AJIM66,6

Dow

nloa

ded

by U

nive

rsity

of

Tex

as L

ibra

ries

At 1

1:48

02

Nov

embe

r 20

14 (

PT)

Date(m

onth/day/

year)

Modularity

No.of

clusters

Cluster1

(%)

Cluster2

(%)

Cluster3

(%)

Cluster4

(%)

Cluster5

(%)

Cluster6

(%)

Cluster7

(%)

Liberal

(%)

Conservativ

e(%

)

11/26/11

0.172

636.67

18.62

17.58

10.68

8.98

1.04

–82.89

10.68

11/30/11

0.223

432.28

25.48

23.75

13.69

––

–69.72

25.48

12/14/11

0.222

433.85

29.95

26.3

3.12

––

–56.25

33.85

12/18/11

0.179

626.91

18.6

1412.47

11.82

10.94

–67.83

26.91

12/19/11

0.209

438.03

22.14

17.6

17.41

––

–73.04

22.14

12/24/11

0.156

635.62

21.14

12.34

11.48

11.37

3.54

–91.95

3.54

12/25/11

0.184

443.77

21.22

19.98

11.8

––

–84.97

11.8

12/26/11

0.216

538.55

19.74

18.91

16.96

3.15

––

75.25

18.91

01/01/12

0.153

431.7

24.56

22.71

13.99

––

–78.97

13.99

01/14/12

0.170

532.97

26.49

265.98

5.08

––

90.54

5.98

01/15/12

0.145

632.6

25.64

13.59

8.73

7.07

5.64

–84.54

8.73

01/21/12

0.170

620.08

16.94

16.45

15.87

14.79

10.87

–78.55

16.45

01/22/12

0.200

522.43

21.11

20.36

19.49

11.4

––

73.68

21.11

01/27/12

0.222

435.1

23.62

18.94

18.51

––

–72.55

23.62

01/29/12

0.254

628.59

21.17

15.88

12.61

11.55

1.44

–70.07

21.17

02/09/12

0.264

434.96

30.71

22.04

5.03

––

–52.75

34.96

03/10/12

0.169

633.66

18.14

16.88

14.08

10.56

2.35

–82.76

10.56

03/18/12

0.139

723.73

19.96

18.57

11.92

9.93

7.35

4.27

83.81

11.92

03/24/12

0.189

548.64

19.1

18.34

6.49

5.55

––

91.63

6.49

03/30/12

0.229

440.22

31.84

24.85

1.8

––

–65.07

31.84

Table IV.Modularity analysis of

clusters

595

By Koreantwitter users

Dow

nloa

ded

by U

nive

rsity

of

Tex

as L

ibra

ries

At 1

1:48

02

Nov

embe

r 20

14 (

PT)

like-mined individuals in terms of political viewpoints on the internet (Sunstein, 2007),particularly through SNSs.

Discussion and conclusionsTwitter has clearly become a major part of global cyberspace. An increasing number ofresearchers have focused on the effects of Twitter and its role as a forum for politicaldiscourse. Although many studies have examined the political use of Twitter, relativelyfew have investigated its use in Asian contexts. As discussed earlier, Korea providesresearchers with a unique setting for examining how political discourse is shapedand disseminated through the internet, particularly through Twitter (Choi andPark, 2014).

In this study, a Twitter network of users’ discussions on a Korean president wasmapped as a way to identify opinion leaders in the network. Based on network density,the strength of connections between users in each cluster was examined. This study’sapproach is expected to complement surveys or experimental studies examining opinionleadership and the homophily phenomenon. The study contributes to the literatureon political communication by increasing methodological diversity and providing a cross-national comparison of the role of Twitter in political discourse.

The results for RQ1 indicate that a few liberal Twitter users influenced informationflow through their Twitter messages within the examined discussion network. In thepresent study, opinion leaders were already influential in the offline world, and this wasthe case regardless of the social issue being discussed. Noteworthy is that Himelboimet al. (2013) found that “more specific topics of controversy had both conservativeand liberal clusters, while in broader topics, dominant clusters reflected conservativesentiment” (p. 154). In the present study, liberal-minded opinion leaders on Twitterdominated discussions on Myung-Bak Lee.

The results for RQ2 indicate that the political discourse on Twitter was fragmentedinto subgroups of like-minded users, which provides support for the homophily thesis.In particular, the results of the SNA suggest that the homophily thesis can be applied to

Liberal Lib. - Cons.Conservative

0.180

0.120

0.100

0.080

0.060

0.040

0.020

0.0002011-11-26 2011-12-26 2012-01-26 2012-03-262012-02-26

0.160

0.140

Figure 2.Longitudinal density forliberal and conservativeclusters and betweenclusters

596

AJIM66,6

Dow

nloa

ded

by U

nive

rsity

of

Tex

as L

ibra

ries

At 1

1:48

02

Nov

embe

r 20

14 (

PT)

Twitter, a microblogging site. However, the analysis focused only on Twitterdiscussions about the former Korean president. Although the results clearly indicatepolitical polarization for the data, it remains unclear whether the results can begeneralized to other topics and to what extent. In this regard, future research shouldapply this study’s approach to other social contexts.

This study has some limitations. The results are not based on an exhaustive dataset. Although the data were collected from November 1, 2011, to April 20, 2012, not allwere included in the analysis for technical reasons. In addition, data collection wasbased on those tweets including the full name of the former president, namely “ .”This might have excluded tweets referring to him with alternative terms such as hisinitial “MB,” although all available tweets provided by the Twitter API during theanalysis period were collected. Given that the data were collected over a five-monthperiod, the results are expected to sufficiently capture political discussions amongKorean Twitter users during normal days. One particularly important limitation of thestudy is that egocentric snowball sampling had to be employed to infer Twitter users’political views, which were drawn from political views of adjacent central nodes withthe highest indegree centrality within each cluster. Therefore, the results of this casestudy should be interpreted with caution. Given the large number of Twitter usersincluded in the analysis, it was not possible to manually code all Twitter users’ politicalviews. That is, political views were manually coded only for systematically selectedcentral nodes showing the highest indegree centrality. As noted earlier, however, asemantic analysis was complementarily conducted to overcome the potential weaknessof inferring peripheral nodes’ political views. The results show strong correlationsbetween tweets by core nodes (10 percent) and those by peripheral ones (90 percent) interms of word frequency (see Table III). Future research should provide a more in-depthempirical understanding through a more extensive content analysis of actual tweets.Finally, it should be noted that three types of relationships (“followings,” “mentions,”and “retweets”) were combined to address the proposed research questions. Given thatsome studies have reported differences between mention and retweet networks, theresults of this study need to be verified by focusing on each relationship separately.

Despite these limitations, the results have important implications. Most of theopinion leaders identified in the study held liberal political views, and liberalTwitter users dominated the discussion network (see Table IV). In addition, the resultsdemonstrate the presence of the homophily phenomenon within the Korean Twitternetwork, implying that opinion leaders’ influence within the Twitter network may belimited to other users sharing the same political views. Further, the results suggest thatpolitical views of opinion leaders on Twitter may be skewed toward a particularpolitical stance without necessarily representing the opinion of the general public,which can be detrimental to the democratic process. Given that each country has itsown distinct media environment, a comparative analysis of the formation of opinionleadership and the homophily phenomenon on Twitter should provide a betterunderstanding of political microblogging activities and their societal implications forthe broader public.

Notes1. See www.statista.com/statistics/284473/south-korea-social-network-penetration/

2. Sociometry refers to “a science of measuring the degree of interrelatedness among people andobjects on a given dimension […] socio metric studies typically measure and observe

597

By Koreantwitter users

Dow

nloa

ded

by U

nive

rsity

of

Tex

as L

ibra

ries

At 1

1:48

02

Nov

embe

r 20

14 (

PT)

attractions/repulsions within a given group, such as family, social, work, and community”(Barnett, 2011, p. 797).

3. During the data collection period, Myung-Bak Lee was the incumbent president, but as ofFebruary 2013, Geun-hye Park became the eleventh and current president of Korea (Park,2014). In the collection of tweets, efforts were made to avoid using topics that coulddisproportionably attract one group of users with the same political views. In addition,the use of time-sensitive topics potentially not reflecting the characteristics of Korean Twitternetworks was avoided. Based on these factors, the longitudinal distribution of tweetscontaining the president’s name was examined.

4. For binary data, network density is calculated as (the number of relationships in thenetwork)/(the number of nodes) * (the number of nodes – 1) (Wasserman and Faust, 1994) forinner-groups. For valued data, network density is defined as the sum of values for all presentrelationships divided by the number of all possible ones.

5. UCINET allows researchers to display “the block model network, the set of ties in the originalmatrix but with the people (nodes) sorted according to their position. One can easily see ifthere are links within and between positions: blocks with lots of zeros represent noconnection between positions, whereas blocks with lots of ones indicate connections betweenpositions […]. If block densities are larger than the overall network density, there is aconnection or link between these blocks” (Valente, 2010, pp. 117-119).

References

An, J., Cha, M., Gummadi, K. and Crowcroft, J. (2011), “Media landscape in twitter: a world of newconventions and political diversity”, Proceedings of the Fifth International AAAIConference onWeblogs and Social Media in Barcelona, Spain, The AAAI Press, Menlo Park,pp. 18-25.

An, J., Quercia, D. and Crowcroft, J. (2013), “Fragmented social media: a look into selectiveexposure to political news”, Proceedings of the 22nd International Conference on WorldWide Web Companion in Rio de Janeiro, Brazil, 2013, International World Wide WebConferences Steering Committee, Republic and Canton of Geneva, Switzerland, pp. 51-52.

Barberá, P. (2012), “A new measure of party identification in twitter. Evidence from Spain”,paper presented at the second European Political Science Association Conference (EPSA),June 21-23, Berlin, available at: https://files.nyu.edu/pba220/public/barbera_epsa2012.pdf(accessed July 10, 2013).

Barnett, G.A. (2011), Encyclopedia of Social Networks, Sage Publications, Los Angeles, CA.

Bruns, A. (2005), Gatewatching: Collaborative Online News Production, Peter Lang, New York, NY.

Bruns, A. and Burgess, J. (2011), “New methodologies for researching news discussion onTwitter”, paper presented at the Future of Journalism 2011, Cardiff, available at: http://eprints.qut.edu.au/46330 (accessed July 10, 2013).

Cha, M., Haddadi, H., Benevenuto, F. and Gummadi, K.P. (2010), “Measuring user influence intwitter: the million follower fallacy”, Proceedings of the Fourth International AAAIConference on Weblogs and Social Media in Washington DC, 2010, The AAAI Press, MenloPark, pp. 10-17.

Chang, W. and Park, H.W. (2012), “The network structure of the Korean blogosphere”, Journal ofComputer-Mediated Communication, Vol. 17 No. 2, pp. 216-230.

Cho, S.E. and Park, H.W. (2012), “Government organizations’ innovative use of the internet: thecase of the twitter activity of South Korea’s ministry for food, agriculture, forestry andfisheries”, Scientometrics, Vol. 90 No. 1, pp. 9-23.

598

AJIM66,6

Dow

nloa

ded

by U

nive

rsity

of

Tex

as L

ibra

ries

At 1

1:48

02

Nov

embe

r 20

14 (

PT)

Cho, S.E. and Park, H.W. (2013), “A qualitative analysis of cross-cultural new media research:SNS use in Asia and the west”, Quality and Quantity, Vol. 47 No. 4, pp. 2319-2330.

Choi, S. and Park, H.W. (2014), “An exploratory approach to a twitter-based community centeredon a political goal in South Korea: who organized it, what they shared, and how theyacted”, New Media and Society, Vol. 16 No. 1, pp. 129-148.

Chong, D. and Druckman, J. (2007), “Framing theory”, Annual Review of Political Science, Vol. 10,pp. 103-126.

Chou, W., Hunt, Y., Beckjord, E., Moser, R. and Hesse, B. (2009), “Social media use in the UnitedStates: implications for health communication”, Journal of Medical Internet Research,Vol. 11 No. 4, doi: 10.2196/jmir.1249.

Chung, C.J., Cho, S.E. and Park, H.W. (2014), “SNS use by the Korean government: a case ofMe2Day”, Asian Journal of Communication, Vol. 24 No. 1, pp. 25-41.

Conover, M.D., Gonçalves, B., Ratkiewicz, J., Flammini, A. and Menczer, F. (2011), “Predicting thepolitical alignment of twitter users”, Privacy, Security, Risk and Trust (Passat), 2011 IEEEThird International Conference on and 2011 IEEE Third International Conference on SocialComputing (SocialCom) in Boston, MA, pp. 192-199.

Conover, M.D., Ratkiewicz, J., Francisco, M., Gonçalves, B., Flammini, A. and Menczer, F. (2011),“Political polarization on twitter”, Proceedings of the 5th International Conference onWeblogs and Social Media in Barcelona, Spain, The AAAI Press, Menlo Park, pp. 89-96.

Denton, R. and Kuypers, J. (2008), Politics and Communication in America: Campaigns, Media,and Governing in the 21st Century, Waveland Press, Long Grove, IL.

Ellison, N., Steinfield, C. and Lampe, C. (2007), “The benefits of facebook ‘friends’: social capitaland college students’ use of online social network sites”, Journal of Computer-MediatedCommunication, Vol. 12 No. 4, pp. 1143-1168.

Fischer, P., Greitemeyer, T. and Frey, D. (2008), “Self-regulation and selective exposure: theimpact of depleted self-regulation resources on confirmatory information processing”,Journal of Personality and Social Psychology, Vol. 94 No. 3, pp. 382-395.

Gil de Zúñiga, H., Puig, E. and Rojas, H. (2009), “Weblogs traditional sources online and politicalparticipation: an assessment of how the internet is changing the political environment”,New Media and Society, Vol. 11 No. 4, pp. 553-574.

Golbeck, J. and Hansen, D. (2011), “Computing political preference among twitter followers”,Proceedings of the SIGCHI Conference on Human Factors in Computing Systems in NewYork, USA, ACM, New York, pp. 1105-1108.

Golbeck, J., Grimes, J.M. and Rogers, A. (2010), “Twitter use by the US congress”, Journal of theAmerican Society for Information Science and Technology, Vol. 61 No. 8, pp. 1612-1621.

Gruzd, A. (2012), “Investing political polarization on twitter: a Canadian perspective”, paperpresented at Internet, Politics, Policy 2012, September 20-21, Oxford, available at: http://microsites.oii.ox.ac.uk/ipp2012/papers (accessed July 10, 2013).

Gruzd, A., Wellman, B. and Takhteyev, Y. (2011), “Imagining twitter as an imagined community”,American Behavioral Scientist, Vol. 55 No. 10, pp. 1294-1318.

Gueorguieva, V. (2007), “Voters, myspace, and youtube: the impact of alternative communicationchannels on the 2006 election cycle and beyond”, Social Science Computer Review, Vol. 26No. 3, pp. 288-300.

Harrow, S. and Harp, D. (2012), “Collective action on the web”, Information, Communication &Society, Vol. 15 No. 2, pp. 196-216.

Haythornthwaite, C. (1996), “Social network analysis: an approach and technique for the study ofinformation exchange”, Library & Information Science Research, Vol. 18 No. 4, pp. 323-342.

599

By Koreantwitter users

Dow

nloa

ded

by U

nive

rsity

of

Tex

as L

ibra

ries

At 1

1:48

02

Nov

embe

r 20

14 (

PT)

Himelboim, I., McCreery, S. and Smith, M. (2013), “Birds of a feather tweet together: integratingnetwork and content analyses to examine cross-ideology exposure on twitter”, Journal ofComputer-Mediated Communication, Vol. 18 No. 2, pp. 154-174.

Howard, P.N. and Hussain, M.M. (2013), Democracy’s Fourth Wave? Digital Media and the ArabSpring, Oxford University Press, Oxford.

Hsu, C.L. and Park, H.W. (2011), “Sociology of hyperlink networks of web 1.0, web 2.0, and twitter:a case study of South Korea”, Social Science Computer Review, Vol. 29 No. 3, pp. 354-368.

Hsu, C.L. and Park, H.W. (2012), “Mapping online social networks of Korean politicians”,Government Information Quarterly, Vol. 29 No. 2, pp. 169-181.

Hsu, C.-L., Park, S.J. and Park, H.W. (2013), “Political discourse among key twitter users: thecase of Sejong city in South Korea”, Journal of Contemporary Eastern Asia, Vol. 12 No. 1,pp. 65-79.

Kang, K.H. (2013), “Conservatives have disappeared on twitter”,Voices of the People, Vol. 17, June,available at: www.vop.co.kr/A00000645436.html (accessed March 25, 2014).

Katz, E. (1957), “The two-step flow of communication: an up-to-date report on an hypothesis”, ThePublic Opinion Quarterly, Vol. 21 No. 1, pp. 61-78.

Katz, E. and Lazarsfeld, P.F. (1955/2006), Personal Influence: The Part Played By People in theFlow of Mass Communications, Transaction Publishers, New Brunswick, NJ.

Khan, G.F., Yoon, H.Y. and Park, H.W. (2014), “Social media communication strategies ofgovernment agencies: twitter use in Korea and the US”, Asian Journal of Communication,Vol. 24 No. 1, pp. 60-78.

Khan, G.F., Yoon, H.Y., Kim, J.Y. and Park, H.W. (2014), “From e-government to socialgovernment: the case of twitter use by Korea’s central government”, Online InformationReview, Vol. 38 No. 1, pp. 95-113.

Kim, M. and Park, H.W. (2012), “Measuring twitter-based political participation and deliberationin the South Korean context by using social network and triple helix indicators”,Scientometrics, Vol. 90 No. 1, pp. 121-140.

Kim, Y. (2011), “The contribution of social network sites to exposure to political difference:The relationships among SNSs, online political messaging, and exposure to cross-cuttingperspectives”, Computers in Human Behavior, Vol. 27 No. 2, pp. 971-977.

King, A., Orlando, F. and Sparks, D. (2011), “Ideological extremity and primary success: a socialnetwork approach”, paper presented at the Annual Meeting of the Midwest Political ScienceAssociation Conference (MPSA), March 31-April 2, Chicago, IL, available at: http://sites.duke.edu/msufjo/files/2012/06/King-Orlando-Sparks-Ideological-Extremity-and-Primary-Success.pdf (accessed July 10, 2013).

Krippendorff, K. (2004), “Reliability in content analysis”, Human Communication Research,Vol. 30 No. 3, pp. 411-433.

Kushin, M. and Yamamoto, M. (2010), “Did social media really matter? College students’ use ofonline media and political decision making in the 2008 election”,Mass Communication andSociety, Vol. 13 No. 5, pp. 608-630.

Leavitt, A., Burchard, E., Fisher, D. and Gilbert, S. (2009), “The influentials: new approaches foranalyzing influence on twitter”, available at: www.webecologyproject.org/wp-content/uploads/2009/09/influence-report-final.pdf (accessed July 10, 2013).

Lee, N.Y. and Kim, Y. (2014), “The spiral of silence and journalists’ outspokenness on twitter”,Asian Journal of Communication, Vol. 24 No. 3, pp. 262-278.

Lee, W.T., Cha, M.Y. and Yang, H.R. (2011), “The network characteristic of social mediainfluentials: focusing on South Korea’s twitter networks”, Journal of CommunicationResearch, Vol. 48 No. 2, pp. 44-79.

600

AJIM66,6

Dow

nloa

ded

by U

nive

rsity

of

Tex

as L

ibra

ries

At 1

1:48

02

Nov

embe

r 20

14 (

PT)

Monge, P. and Contractor, N. (2003), Theories of Communication Networks, Oxford UniversityPress, Oxford.

Mutz, D.C. (2002), “Cross-cutting social networks: testing democratic theory in practice”, AmericanPolitical Science Review, Vol. 96 No. 1, pp. 111-126.

Mutz, D.C. (2006), Hearing the Other Side: Deliberative Versus Participatory Democracy,Cambridge University Press, New York, NY.

Newman, M.E.J. (2006), “Modularity and community structure in networks”, Proceedings ofthe National Academy of Sciences of the United States of America, Vol. 103 No. 23,pp. 8577-8582.

Owen, S. (2008), “Citizens, media use social media to monitor election”, available at: www.pbs.org/mediashift/2008/11/citizens-media-use-social-media-to-monitor-election308.html (accessedJuly 10, 2013).

Park, H.W. (2014), “Mapping election campaigns through negative entropy: triple and quadruplehelix approach to South Korea’s 2012 presidential election”, Scientometrics, Vol. 99 No. 1,pp. 187-197.

Park, H.W. and Thelwall, M. (2008), “Developing network indicators for ideological landscapesfrom the political blogosphere in South Korea”, Journal of Computer MediatedCommunication, Vol. 13 No. 4, pp. 856-879.

Pew Research Center (2011), “Twitter and the campaign: how the discussion on twitter varies fromblogs and news coverage and Ron Paul’s twitter triumph”, available at: www.journalism.org/analysis_report/twitter_and_campaign?src¼ prc-headline (accessed July 10, 2013).

Rainie, L. and Smith, A. (2012), “Social networking sites and politics”, available at: http://pewinternet.org/~/media//Files/Reports/2012/PIP_SNS_and_politics.pdf (accessed July 10,2013).

Rho, W.-J. (2014), “Triple helix for social innovation: the saemaul undong for eradicatingpoverty”, Journal of Contemporary Eastern Asia, Vol. 13 No. 1, pp. 39-55.

Rogers, E. (2003), Diffusion of Innovations, 5th ed., Free Press, New York, NY.

Small, T.A. (2011), “What the hashtag? A content analysis of Canadian politics on twitter”,Information, Communication & Society, Vol. 14 No. 6, pp. 872-895.

Smith, A. (2011), “22% of online Americans used social networking or Twitter for politics in 2010campaign”, available at: http://pewinternet.org/~/media/Files/Reports/2011/PIP-Social-Media-and-2010-Election.pdf (accessed July 10, 2013).

Smith, A. and Rainie, L. (2008), “The internet and the 2008 election”, available at: www.pewinternet.org/~/media//Files/Reports/2008/PIP_2008_election.pdf (accessed July 10,2013).

Stroud, N. (2007), “Media effects, selective exposure, and fahrenheit 9/11”, Political Communication,Vol. 24 No. 4, pp. 415-432.

Stroud, N. (2010), “Polarization and partisan selective exposure”, Journal of Communication,Vol. 60 No. 3, pp. 556-576.

Sunstein, C. (2007), Republic.com 2.0, Princeton University Press, Princeton, NJ.

Takhteyev, Y., Gruzd, A. andWellman, B. (2012), “Geography of twitter networks”, Social Networks,Vol. 34 No. 1, pp. 73-81.

Valente, T.W. (2010), Social Networks and Health: Models, Methods, and Applications, OxfordUniversity, New York, NY.

Valente, T.W. and Pumpuang, P. (2007), “Identifying opinion leaders to promote behaviorchange”, Health Education & Behavior, Vol. 34 No. 6, pp. 881-896.

601

By Koreantwitter users

Dow

nloa

ded

by U

nive

rsity

of

Tex

as L

ibra

ries

At 1

1:48

02

Nov

embe

r 20

14 (

PT)

Valenzuela, S., Park, N. and Kee, K. (2009), “Is there social capital in a social network site?Facebook use and college students’ life satisfaction, trust, and participation”, Journal ofComputer-Mediated Communication, Vol. 14 No. 4, pp. 875-901.

Wasserman, S. and Faust, K. (1994), Social Network Analysis: Methods and Applications,Cambridge University Press, Cambridge.

Weimann, G., Tustin, D.H., van Vuuren, D. and Joubert, J.P.R. (2007), “Looking for opinion leaders:traditional vs. modern measures in traditional societies”, International Journal of PublicOpinion Research, Vol. 19 No. 2, pp. 173-190.

Wu, S., Hofman, J.M., Mason, W.A. and Watts, D.J. (2011), “Who says what to whom on twitter”,Proceedings of the 20th International Conference on World Wide Web in Hyderabad, India,ACM, New York, NY, pp. 705-714.

Xu, W.W., Sang, Y., Blasiola, S. and Park, H.W. (2014), “Predicting opinion leaders in twitteractivism networks: the case of the Wisconsin recall election”, American BehavioralScientist, Vol. 58 No. 10, pp. 1278-1293

Yoon, H.Y. and Park, H.W. (2014), “Strategies affecting twitter-based networking pattern of SouthKorean politicians: social network analysis and exponential random graph model”, Quality& Quantity, Vol. 48 No. 1, pp. 409-423.

Zhang, W., Johonson, T., Seltzer, T. and Bichard, S. (2010), “The revolution will be networked: theinfluence of social networking sites on political attitudes and behavior”, Social ScienceComputer Review, Vol. 28 No. 1, pp. 75-92.

About the authorsMyunggoon Choi is a Master’s Degree candidate in the Department of Interaction Science at theSungkyunkwan University. His research interests focus on user experience, social networkanalysis, human-technology interactions, and webometrics. He is also a user data analyst ofCoupang, an e-commerce company in South Korea. He was a Research Assistant at the WorldClass University Webometrics Institute.

Yoonmo Sang is a Doctoral Candidate in the Department of Radio-Television-Film at theUniversity of Texas-Austin. His research interests lie at the intersection of new media and law.In particular, he is most interested in the ways in which Internet intermediaries deal withsocial controversies resulting from rapid technological changes, such as online libel, copyrightinfringement, and invasion of privacy, as well as the impact of new media on politics. Hisscholarly writing has appeared in Journal of Media Law and Ethics, Journal of Medical Systems,Computers in Human Behavior, and American Behavioral Scientist.

Professor Han Woo Park is Full Professor in the Department of Media and Communicationat the YeungNam University, South Korea. His research focuses on the use of new digitaltechnologies in extending social and semantic networks and the role of communication inscientific, technical, innovative, and governmental activities. He has been a Research Associate atthe Royal Netherlands Academy, a visiting scholar at the Oxford Internet Institute, and a directorof the World Class University Webometrics Institute. He is currently the Director of the CyberEmotions Research Center and Asia Triple Helix Society. He has founded a prestigiousconference on Network Sciences in Asia, called DISC (Daegu Gyeongbuk International SocialNetwork Conference, http://asia-triplehelix.org/DISC2013), He sits on the editorial boards ofprestigious journals including Journal of Contemporary Eastern Asia, Scientometrics, Quality &Quantity, Knowledge Economy, and Big Data & Society. Professor Han Woo Park is thecorresponding author and can be contacted at: [email protected]

To purchase reprints of this article please e-mail: [email protected] visit our web site for further details: www.emeraldinsight.com/reprints

602

AJIM66,6

Dow

nloa

ded

by U

nive

rsity

of

Tex

as L

ibra

ries

At 1

1:48

02

Nov

embe

r 20

14 (

PT)