Content uploaded by Raquel Recuero
Author content
All content in this area was uploaded by Raquel Recuero on Jan 30, 2019
Content may be subject to copyright.
Content uploaded by Raquel Recuero
Author content
All content in this area was uploaded by Raquel Recuero on Jan 30, 2019
Content may be subject to copyright.
Pre-print version of: Felipe Bonow Soares, Raquel Recuero, and Gabriela Zago. 2018. Influencers in Polarized Political Networks on Twitter.
In Proceedings of the 9th International Conference on Social Media and Society (SMSociety '18). ACM, New York, NY, USA, 168-177. DOI:
https://doi.org/10.1145/3217804.3217909
Influencers in Polarized Political Networks on Twitter
Felipe Bonow Soares
Federal University of Rio Grande do Sul
Faculdade de Biblioteconomia e
Comunicação
R. Ramiro Barcelos, 2705
Prédio 22201
Porto Alegre, RS, Brazil
felipebsoares@hotmail.com
Raquel Recuero
Federal University of Pelotas
Federal University of Rio Grande do Sul
Faculdade de Biblioteconomia e
Comunicação
R. Ramiro Barcelos, 2705
Prédio 22201
Porto Alegre, RS, Brazil
raquelrecuero@gmail.com
Gabriela Zago
University of Oregon
Allen Hall, 1275
Eugene, OR, USA
gabrielaz@gmail.com
ABSTRACT
This study aims to identify influencers roles in political
conversations on Twitter and their effect in the online public
sphere. We analyze data from three key days during the
impeachment process of the ex-president of Brazil, Dilma
Rousseff. Using a combination of social network analysis
and qualitative approaches, we discuss these political
influencers and their role in the conversations that took
place. Our main findings are: (1) the presence of highly
modularized networks, where groups in favor or opposing
the impeachment were tightly connected around different
influencers; (2) the identification of three different types of
influencers, the “opinion leaders”, the “informational
influencers” and the “activists” and their different roles in
the influential process and the polarization of the network.
CCS CONCEPTS
• Human-centered computing → Collaborative and social
computing → Collaborative and social computing theory,
concepts and paradigms → Social networks
KEYWORDS
Influencers, polarized networks, public sphere, social
network analysis, social networks
ACM Reference format:
Soares, F. B. Recuero, R. & Zago, G. 2018. Influencers in Polarized
Political Networks on Twitter. In Proceedings of the International
Conference on Social Media & Society, Copenhagen, Denmark
(SMSociety). DOI: 10.1145/3217804.3217909
1 INTRODUCTION
Many studies have discussed influencers on social media in
general [1-5] and in the political context specifically [5].
However, fewer studies focused on the role these influencers
may play in shaping the conversation in the political public
sphere and the effects this may have on young democracies
such as Brazil - which has had a long period of military
dictatorship in the 20th century. It is important to understand
the relationship among political influencers and social media
and how they might affect the public opinion. It is especially
relevant to study it in the Brazilian context, considering the
country’s political turbulence and the upcoming elections in
2018.
In this paper, we seek to identify influencers roles on
political conversations on Twitter and discuss its effect on
the public sphere. For this purpose, we chose to analyze
tweets about the impeachment of the ex-president of Brazil
Dilma Rousseff, which took place in 2016, with data from
three days during different moments of the process. Dilma
Rousseff, a left-wing politician, was impeached on August
31st of 2016, amidst a very intense and polarized political
debate. Those aligned with the right-wing were favorable to
the impeachment and even organized events on key dates.
Those aligned with the left-wing questioned the process,
accusing the politicians enacting a “coup” rather than an
impeachment, claiming the process had nothing to do with
the legal matter at hand. The impeachment process started at
the end of 2015 when the National Congress accepted the
procedure started by three lawyers and supported by groups
politically opposed to Rousseff. The Chamber of Deputies
voted in favor on April 17th 2016 and then, the Federal
__________________________________________
Permission to make digital or hard copies of part or all of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. Copyrights for third-
party components of this work must be honored. For all other uses, contact
the owner/author(s).
SMSociety, July 2018, Copenhagen, Denmark
© 2018 Copyright held by the owner/author(s).
$15.00. DOI: 10.1145/3217804.3217909
SMSociety, July 2018, Copenhagen, Denmark
F. B. Soares, R. Recuero, G. Zago
Senate discussed and accepted the process on May 12th. The
president defended herself personally twice in Federal
Senate sessions (August 29th and August 30th). The Senate
voted in favor of the impeachment on August 31st.
The paper is organized as follows: first, we discuss the
theoretical background with political conversations, public
sphere and polarized groups, followed by the research
questions that guided this study. Next, we present the
methods, followed by a discussion of the results.
2 POLITICAL INFLUENCERS, POLARIZED
GROUPS AND PUBLIC SPHERE
2.1 Opinion leaders and influencers
Dubois and Gaffner [5] define the ability to influence as
“convincing an individual to change his or her opinion,
attitude, and/or behavior” (p. 1261). Their work discusses
two types of influencers: the opinion leader, who uses the
content to influence; and the influential, who uses network
position to spread messages.
The opinion leader theory is linked to the early work of
Lazarsfeld, Berelson and Gaudet [6] and the two-step flow
of communication. In their discussion, personal influence
was studied as a process where some people who received
material directly from mass media would influence other
people. They studied a presidential campaign and found that
a few individuals were mostly responsible for influencing
others politically. The authors called them as “opinion
leaders” - highly political engaged individuals responsible
for influencing those less active. In this approach,
identifying influencers was a matter of asking people who
influenced them and if they thought they influenced others.
At that time, many scholars believed that the formal media
was the most relevant influencer, but Lazarsfeld, Berelson
and Gaudet discovered that people were more likely to be
influenced by informal political conversations with other
people than by the formal media.
Berelson, Lazarsfeld and McPhee [7] also identified a
selectivity process, discovering that individuals tend to
reinforce their own opinion in persuasive processes. This
means that the opinion leaders tend to have stronger
persuasive power in groups where people have similar views
and their influence reinforces their own ideas. The
selectivity process also means that individuals are inclined
to look for opinion leaders that spread opinions equivalent to
their own.
Dubois and Gaffner [5] showed that one can be an
influencer based on one's network position. In social media,
the influential theory uses network structure and information
diffusion in order to identify those “influencers” [1-5],
relying in network analysis and other quantitative
measurements to find these actors. This approach usually
uses large scale social data and quantitative measurements to
identify and understand how network actors influence each
other. In order to identify influencers, Dubois and Gaffner
[5] used a combination of metrics - indegree, eigenvector
centrality (network placement), clustering coefficient (social
embeddedness), knowledge, and interaction (content
analysis) - and found that each group revealed different top
users. When considering network placement, media outlets,
journalists, and politicians are the most influential, whereas
interaction and content show those actors, plus political
commentators and bloggers, as influential. The local level
reveals different actors as likely opinion leaders in their
neighborhoods. The most important conclusion, is that even
though influence is usually studied by looking at number of
followers and/or expertise, other facets of influence can also
be important.
Cha et al. [1] aimed to identify who those influencers on
social media are and found that they are users that strongly
invest in their reputation in a specific area. These users
develop a persuasive power within some groups interested in
a topic where they are recognized as authorities. For
example, a famous singer with millions of followers tends to
have less influence in a political discussion than a person
who invests social capital to be recognized as an expert in
some political matter – even if this user has less followers
than the singer. In a similar sense, Bakshy et al. [3] identify
different types of influencers: users with a high following
count, and “ordinary influencers”, who due to circumstance
exert local influence on others.
Few of these studies analyze the role these different types
of influencers may have in political conversations and in
social media as a public sphere. This is a key point to
understanding how social media may influence democracy,
and the role it has as a public sphere. We posit that it is not
only those with high visibility, or those that have their
message spread among the network are influencers, but also
those with a high activity who might also shape the
conversation by giving visibility and relevance to topics or
opinions.
Studying discussion forums, Graham and Wright [4]
discuss the impact of what they call “superparticipants”:
actors that participate much more than average. They
identify three roles of superparticipants: superposters,
agenda-setters, and facilitators. In their study, while
superposters account for only 0.4% of the users, they made
47% of the posts, indicating intense activity by these users.
Their role was largely positive, helping other users by
summarizing content and organizing the discussion. The
agenda-setters were those responsible for suggesting new
topics for discussion. The facilitators were the ones who
Influencers in Polarized Political Networks on Twitter SMSociety, July 2018, Copenhagen, Denmark
managed the activities on the forums. Graham and Wright
findings were related to different roles each type of
influencer had, similar to what we seek to discuss in our
analysis.
2.2 Public sphere and social capital
Influencers play an important role in shaping the public
sphere. The public sphere is a concept first used by Jürgen
Habermas, related to the public place where discussion about
society by its members could take place. Habermas [8]
initially believed in a public sphere similar to the Athenian
democracy, with the citizens occupying equal hierarchical
positions when deliberating. However, in Between Facts and
Norms, Habermas [9] suggested the existence of some actors
more capable of influencing the discussion in the public
sphere, and thus identifying that not all citizens have the
same power in the discussion. In this second sentiment,
Habermas considered the public sphere as a network with
many arenas connected by a communicational flow that
encourages the debate between groups and individuals with
different ideas. According to Habermas [9], the influencers
in public sphere have a social power in some specialized
areas, such as religious, scientific, and political groups. Their
reputation is constructed based on the other citizens' belief
and they have a strong power for shaping the public opinion
-especially in those groups they have the most influence. A
key concept to understanding the influence based on
reputation, as described by Habermas, is social capital.
Social capital is a concept related to the “social
resources” [10] to which being a member of a group gives
an individual access. According to Coleman [11], social
capital is a structure that facilitates actions within a social
structure and can assume different forms. An important
aspect of online social networks is that they can facilitate
access to forms of social capital that would otherwise be
inaccessible to users [12, 13]. Therefore, social capital is a
key component of influence in social media.
2.3 Public Sphere and Social Media
The public sphere has a unique structure on social media.
Research has showed that on Twitter the structure of the
public sphere is based on groups, similar to Habermas’
arenas, and dependent on bridges for interaction with other
groups. Bastos [14] describe the digital public sphere as
onion rings, many groups connected on their borders. Bruns
and Highfield [15] found what they called sphericules,
connected by some key users capable of carrying messages
from one group to another. Other authors [16, 17] suggest
that because the debate is highly fragmented, Twitter could
not be considered as a large public sphere, but rather a group
of small, clustered and often disconnected groups. It is
exactly this fragmentation of the debates and isolation of
groups that might threaten the existence of an online public
sphere [16]. Papacharissi [18] suggests that it is possible to
differentiate the new public space in social media and a
potential public sphere in these spaces. The author considers
that the public space generated in social media enhances
discussion, but for it to become a public sphere, it is
necessary to also enhance democracy.
The discussion on how a fragmented public sphere may
influence democracy is very important because research has
showed that people tend to reevaluate their own opinions
when in contact with different views [19]. Disagreement is
then essential for better deliberation [19] and also for
democratic debate.
The fragmentation of the public sphere can thus affect
social media’s democratic potential, since people inside
closed groups tend to only get in contact with similar views,
as online social networks tend to follow homophily patterns
[20, 21].
2.4 Echo chambers and political polarization
An important characteristic of political debates is the
possibility of polarization. Sunstein [22] studied polarized
political groups and found what he calls “echo chambers”
The author found out that these polarized groups are often
formed by like-minded individuals that isolate themselves
from other groups with different ideas, which creates echo
chambers within these isolated groups where only one view
is reverberated. This way, the individuals in these echo
chambers only engage in discussions that reinforce their own
points of view.
Social media channels often use algorithms for social
filtering. These tools may also increase the possibility of
polarized groups. Pariser [17] affirms that the algorithmic
influence creates a unique universe of information around
users based on their preferences. He calls this universe "filter
bubble”. When the content is filtered based on these
algorithms, users tend to only access information and
opinions they agree with. According to Pariser, this may
cause a restriction of access to new information, which is a
threat to the individual's full comprehension of society.
The formation of polarized groups has also been found
in several other social media studies. For example, Smith et
al. [23] discuss the "polarized crowd", a social network
based on two polarized groups, which is a common structure
in political networks. These types of groups have also been
found in Brazilian political context [24]. These polarized
groups can be so strong in some contexts that they draw users
with a relatively neutral position into the debate, as can be
seen in the case of some news outlets [25, 26].
SMSociety, July 2018, Copenhagen, Denmark
F. B. Soares, R. Recuero, G. Zago
The influencers and their roles in political discussion and
in the online public sphere might be related to the structure
of the network. The different types of influencers we
described - the opinion leaders, the ordinary influencers and
the three roles of superparticipants - might have their
influence modified in polarized contexts, which we aim to
discuss in our analysis.
2.5 Research questions
In this paper, we combine network measurements with
qualitative analysis in order to identify influencers in
discussions associated with Brazilian political context.
There are some key research questions we follow to guide
our analysis.
We aim to show a connection between the networks we
analyze with the idea of echo chambers and bubbles.
Research has identified the formation of echo chambers in
several contexts, including Brazil [24], where even media
outlets tend to be drawn into a bubble [25, 26].
Consequently, our first question is: (1) Are the political
conversations about the impeachment polarized? The type of
structure we identify in our networks may help understand
the influencer's role in the conversations.
We seek to discuss the influencer's role in these
conversations and aim to identify these influencers based on
social network analysis metrics. Based on these goals our
other research questions are: (2) What types of influencers
are there in these networks? (3) What are the roles they play
in the conversation? (4) How are their positions related to
the polarization (if it exists)? Finally, we seek to analyze: (5)
How the position of the influencers change (or do not
change) during the period of the analysis?
Our findings will be discussed and linked to the idea of
public sphere in social media, so we can analyze possible
implications and the impact influencers actions might have
in political conversations, specifically in the Brazilian
context.
3 METHODS
For this paper, we collected tweets in Portuguese that
contained the word "impeachment" during March-
September of 2016, the period in which the impeachment of
Brazilian ex-president Dilma Rousseff was discussed and
voted by the National Congress. Data collection was done on
a daily basis, once per hour, respecting Twitter's limits, and
using a crawler for collecting tweets from the REST API.
For the proposed analysis, we will use data from three days
selected from this original dataset. The first is April 26th, the
day the commission responsible to analyze the process was
instituted; May 18th, a few days after the Federal Senate
voted favorable for the process to start in the chamber. May
18th was the day the Brazilian Supreme Court notified
Rousseff because she called the process a “coup” (causing
people started discussing it again); and August 30th, the day
before the final vote of the impeachment, when the president
made her last public defense to the Senate. We decided to
analyze three days (and not more) during different moments
of the impeachment process because we aim to observe the
patterns in the networks and in the influencers actions during
the conversations in different contexts of the process. We
believe that three networks in three important parts of the
impeachment conversation are sufficient to establish
patterns.
In order to analyze the structure of these networks, we
will use Social Network Analysis (SNA) [27, 28]. Analyzing
the structure of the networks is important to show the
position of the nodes and how their roles can be perceived
based on this position.
We chose to observe three metrics: indegree, outdegree,
and modularity. Modularity is a metric used to find groups
in social networks. The higher the modularity, the denser the
connections are within a group (cluster or module) and the
less dense the connections are with other groups [29].
Modularity can be used to understand how strong and dense
the groups are within a Twitter network. In this paper, we
used modularity to highlight the polarization of the networks
discussing the impeachment. This polarization may show the
polarized crowds [23] and echo chambers [22].
Indegree is a metric that shows how many connections a
certain node receives in the network. Since a connection in
our network is marked by a mention or a retweet of a node,
actors with high indegree are the ones whose tweets were the
most influential in the discussion. Indegree may also indicate
which nodes were most central for the discussion in each
group (module). We believe that higher indegree will depict
opinion leaders, whose tweets generated retweets and
mentions among their followers and their groups, as defined
by Lazarasfeld, Berelson and Gaudet [6] and Dubois and
Gaffner [5].
Outdegree, on the other hand, show us which nodes cited
and retweeted other nodes. This metric may show us which
nodes are more engaged in propagating certain ideas and also
influencing what will be made visible to the network and
what will not (therefore, having a central role in filtering
tweets). These highly active users may depict a new category
of the superparticipants originally discussed by Graham and
Wright [4], in the context of Twitter. Table 1 provides a
breakdown of the data collected.
Influencers in Polarized Political Networks on Twitter SMSociety, July 2018, Copenhagen, Denmark
Table 1: Networks data
Date
Tweets
Nodes
Modularity
April 26th
41313
22101
0.544
May 18th
17562
13594
0.804
August 30th
17418
11400
0.657
In order to discuss the influencers we will further analyze
the 12 nodes with higher indegree and outdegree
qualitatively based on their followers, account type, the
assumed position and their actions patterns on Twitter. We
decided to analyze 12 nodes because the nodes with higher
outdegree have very similar numbers and past the 12th user
there are many with the same outdegree (thus, we would
have to analyze a different number of nodes for each
network). Our intention is to understand the type of account
of the node with highest indegree and outdegree and what
type of influence it may depict.
4 RESULTS AND DISCUSSION
4.1 Structure of the Networks
All the analyzed networks have high modularity as seen in
Table 1. This indicates the presence of separated modules as
we can see in the images of each network (Figures 1-3). This
structure suggests that the networks have a fragmented
structure [16] with the presence of echo chambers [22].
When we further qualitatively examined the nodes in
each module, we found accounts pro and anti-impeachment
aggregated in each group, further suggesting that each
module represented a polarized position for or against
Rousseff's impeachment. Accordingly, each module appears
to be a bubble [17], having low or no contact with the other
module and containing only nodes with a similar position.
Figure 1: April 26th, 2016.
Figure 2: May 18th, 2016.
Figure 3: August 30th, 2016.
Figures 1 to 3 show the three networks in our dataset. In
each of the images, we can clearly see two main clusters,
each one comprised of users that are against or in favor of
the impeachment. A high modularity implies that those
nodes are more connected within their groups and less
connected to the rest of the network.
4.2 Roles of nodes
By analyzing users with high indegree and outdegree, we
identified the different potential roles users can play in
political conversations on Twitter, compared to the opinion
leaders and superparticipants described in our theoretical
background.
4.2.1 Indegree. Nodes with higher indegree tend to be the
most influential nodes in the conversation [1, 3, 5]. Each
module has its own influencers and and those most
mentioned in one group are often not mentioned in the other.
Therefore, they are only influencers within groups that share
the same the political position. We also found that once
within a module, the position of an influential node does not
change in all three datasets (see Tables 2-4). This aspect of
SMSociety, July 2018, Copenhagen, Denmark
F. B. Soares, R. Recuero, G. Zago
the network is characterized by the formation of echo
chambers [22], as discussed in our theoretical discussion.
Also, these users tend to have a higher presence because of
retweets. The data suggests that the role of these users in the
group is one of "opinion leaders" [1, 5, 6], with other users
replicating their political tweets through the network.
It is important to highlight that the emergence of echo
chambers does not seem to be only due to the action of
algorithms, as suggests Pariser [17]. Users also seem to
contribute to these echo chambers through retweeting and
mentioning other users. Data suggest that individuals choose
to reinforce their views by retweeting and mentioning like-
minded influencers, actively forming echo chambers and
polarized crowd network structure based on homophily
patterns. The action of looking for opinion leaders with
similar views is not exclusive of social media - Berelson,
Lazarsfeld and McPhee [7] have originally identified this
selectivity process in the 1960s – but social media offers
access to a high volume of like-minded influencers [17, 22].
News outlets also seem to be drawn into specific groups,
even when they do not clearly state their political position.
Among those nodes with higher indegree (Tables 2 to 4),
we found many politicians, journalists, and media outlets –
similar to what was described by Dubois and Gaffner [5].
Most of these accounts have a high number of followers.
Interestingly, most media outlets were only influential
amongst users who were favorable to the impeachment. The
group against the impeachment ignored traditional media
and were instead influenced by other more off mainstream,
alternative outlets and independent media. It also suggests
that the narrative of the process as presented by the
traditional media outlets and by the official profile of
Brazilian Federal Senate (also always inside the pro group)
was seen as biased towards the approval of the impeachment
by the anti-impeachment.
While the traditional media outlets and other institutions,
such as the Brazilian Federal Senate, did not assume a
political position on the matter, many of the nodes with
higher indegree did. The social movement @mblivre, always
inside the pro group, maintained a strongly critical position
against Rousseff at all times. Similarly, the accounts of the
blogs @blogdopim and @o_antagonista, were also clearly
favorable to impeachment and were always inside the pro
module. Some politicians who defended the impeachment
also appear in the pro module, as @senadorcaiado and
@anaamelialemos. Some media collectives - @midianinja
and @j_livres - produced anti-impeachment messages and
were part of the left-wing module (against the
impeachment). Some foreign journalists with more
opinionated messages against the impeachment -
@georgmarques, @ggreenwald and @alexcuadros - were
also identified with the anti-impeachment cluster.
Based on this data, we identified two types of users with
high indegree: those with a relatively neutral position whose
messages were associated to one module position, and did
not reach the other module; and those with a clear position
whose messages were only spread within the like-minded
module. The users with an assumed position were expected
to be inside only one module. However, even those that did
not assume any position, such as the Brazilian Federal
Senate and some news outlets, were drawn into the groups
based on the general users perceptions of their messages.
The users that assumed a position shape what we believe
is the equivalent of an opinion leader [5, 6]. The message of
these opinion leaders influenced other users inside the same
module. The other users then reproduce the opinion leaders
messages to reinforce their own position.
The users with a neutral position, but also inside the
modules tend to have a different influence. They do not
guide other users positions like the opinion leaders do. The
users that did not assume any position are news outlet and
government institutions, so we decided to call they as
“informational influencers”. They also have a central
position in the network, but their force is related to spreading
information.
Both the opinion leaders and informational influencers
are users with many followers. Two of the informational
influencers are the nodes with most followers among those
with higher indegree. This may indicate that visibility is
important for their influence, since it is based on the
spreading of information. Among the opinion leaders, some
nodes, for example @dilmabr, also have many followers,
suggesting that also their political background might be
important for their position as opinion leaders. Some with
less followers, for example @georgmarques, @vinncent and
@alexcruadros, also appears as opinion leaders. These are
more similar to the contextual influencers founded by
Bakshy [3] and the influencers related with a specific topic
founded by Cha et al. [1].
The nodes with higher indegree (Tables 2-4) are divided
in opinion leaders (OL) and informational influencers (II)
and also emphasized if they were inside the pro or anti
impeachment module.
Table 2: Indegree April 26th, 2016
User
Indegree
Followers
Account type
Inf. type
Cluster
anastasia
1.274
25.266
Politician
OL
Pro
g1
652
8.109.743
News outlet
II
Pro
georgmarques
629
12.055
Journalist
OL
Anti
senadofederal
623
328.623
Government
II
Pro
blogdopim
509
94.826
Blog
OL
Pro
Influencers in Polarized Political Networks on Twitter SMSociety, July 2018, Copenhagen, Denmark
Table 2: Indegree April 26th, 2016
User
Indegree
Followers
Account type
Inf. type
Cluster
mudamais
507
36.870
Blog
OL
Anti
dilmabr
496
4.568.369
Politician
OL
Anti
veja
482
7.035.893
News outlet
II
Pro
senadorcaiado
451
142.609
Politician
OL
Pro
telesurtv
409
1.168.958
News outlet
Anti
midianinja
403
62.582
Media
collective
OL
Anti
mblivre
375
34.163
Social
movement
OL
Pro
Table 3: Indegree May 18th, 2016
User
Indegree
Followers
Account type
Inf. type
Cluster
dilmabr
452
4.609.179
Politician
OL
Anti
georgmarques
376
14.781
Journalist
OL
Anti
anaamelialemos
313
34.523
Politician
OL
Pro
vinncent
293
12.474
Journalist
OL
Anti
jotainfo
248
44.335
News outlet
OL
Pro
g1
242
8.197.564
News outlet
II
Pro
alexcuadros
229
7.220
Journalist
OL
Anti
ggreenwald
226
678.961
Journalist
OL
Anti
millylacombe
214
24.410
Journalist
OL
Anti
newyorker
207
6.358.129
News outlet
II
Anti
micheltemer
182
568.512
Politician
OL
Anti
senadofederal
165
342.830
Government
II
Pro
Table 4: Indegree August 30th, 2016
User
Indegree
Followers
Account type
Inf. Type
Cluster
o_antagonista
615
240.360
Blog
OL
Pro
Blogdopim
332
225.672
Blog
OL
Pro
Mblivre
246
42.732
Social
movement
OL
Pro
senadofederal
233
394.876
Government
II
Pro
ggreenwald
163
726.933
News outlet
OL
Anti
requiaopmdb
157
102.142
Politician
OL
Anti
theinterceptbr
153
10.023
Independent
media
OL
Anti
cartacapital
151
1.335.317
News outlet
OL
Anti
andreiasadi
138
69.084
Journalist
OL
Pro
theintercept
134
278.024
Independent
media
OL
Anti
anaamelialemos
132
44.281
Politician
OL
Pro
j_livres
130
76.701
Media
collective
OL
Anti
4.2.2. Outdegree. The outdegree indicates how many
connections one user sent to other users through retweets and
mentions. Therefore, if one particular user retweeted or
mentioned five others in his/her tweets on a particular day,
that means he/she has an outdegree of 5.
Some users have very high outdegree - including users
with 50+ edges. This is a high number of edges, especially
considering each network is comprised of a single day of
tweets. These users also have an overall high number of
tweets, with up to 700,000 total tweets (which means that the
user posts an average of 100 tweets a day), and most of their
activity consists of retweets at other accounts (including
those nodes that appear among the users with the highest
indegree). Some of these characteristics are similar to the
ones found in automated accounts [30]. Further, two of the
higher outdegree accounts - @beijopai and @gervdav - were
suspended and do not exist anymore, also suggesting the
possibility of the existence of automated accounts.
We observed each account of the users with the top 12
higher outdegree in each network. We found that all these
accounts share an "activist" view, where they clearly state
their political position in their account (on their bio,
username, profile picture or in their tweets). These users
clearly made an effort in spreading tweets with political
views similarly to theirs. They seem to play a strong role in
spreading the positions of other like-minded users (mostly
through retweets) within each module. Most accounts seem
to depict only a single user and not a collective of users, such
as a media outlet.
When qualitatively analyzing their Twitter profiles, we
found that some users identify themselves as aligning with
right or left parties, which is a strong indication of an activist
profile. Some users do not have any profile description at all,
but they retweet strong political views. It means they are
very engaged in spreading their views by retweeting or
mentioning other users (mostly retweeting). The majority of
these accounts has a lower number of followers than the ones
with higher indegree, as can be seen in Tables 5 to 7.
Table 5: Outdegree April 26th, 2016
User
Outdegree
Followers
Cluster
beijopai
81
455
Pro impeachment
isaceressuelo
58
800
Pro impeachment
troca_de_canal
58
13.163
Pro impeachment
roy_crabbs
57
37
Pro impeachment
diraholanda
49
12.274
Anti impeachment
fernanda281327
47
193
Pro impeachment
ezequiasns
43
25.845
Pro impeachment
daciradcc
42
842
Pro impeachment
pauloro49195361
42
724
Pro impeachment
pcampos1122
41
518
Pro impeachment
SMSociety, July 2018, Copenhagen, Denmark
F. B. Soares, R. Recuero, G. Zago
Table 5: Outdegree April 26th, 2016
User
Outdegree
Followers
Cluster
kitri1
41
4.354
Pro impeachment
moema4
41
612
Anti impeachment
Table 6: Outdegree May 18th, 2016
User
Outdegree
Followers
Cluster
fernandocesar77
14
2.045
Anti impeachment
gervdav
13
171
Pro impeachment
rosaacl
13
3.794
Pro impeachment
egolpesim
13
319
Anti impeachment
kitri1
12
4.429
Anti impeachment
vladdico
12
605
Anti impeachment
beijopai
12
700
Pro impeachment
leleabreuv
12
1353
Anti impeachment
cassunungarosa
12
304
Anti impeachment
izidiooliver
12
616
Anti impeachment
marcossedassari
12
313
Pro impeachment
diegocorrado
12
652
Anti impeachment
Table 7: Outdegree August 30th, 2016
User
Outdegree
Followers
Cluster
ccris21
39
621
Pro impeachment
gervdav
36
171
Pro impeachment
letrin45
28
11.566
Pro impeachment
seujoca
26
9.608
Pro impeachment
vivianerecife
26
108
Pro impeachment
politicaesaude
25
8.017
Pro impeachment
pauloro49195361
24
1.832
Pro impeachment
guta5300
24
752
Pro impeachment
walderrocha
24
25
Pro impeachment
monica_evelise
21
14
Pro impeachment
miriamdellamora
21
584
Pro impeachment
fpetralha
21
534
Pro impeachment
We observed that nodes with higher outdegree were
normally inside the same module on the three networks. For
example, @beijopai, @gerdav and @pauloro49195361,
appear more than once in the pro-impeachment cluster. Most
of the users with higher outdegree were identified with the
pro-impeachment group. The exception is the dataset from
May 18th that there were more anti than pro among the 12
with higher outdegree. On August 30th, when the possibility
of the impeachment was stronger, there are only nodes inside
the pro-impeachment, showing that the proximity of the final
voting made the pro-activist more active and probably more
confident of the outcome.
The nodes with high outdegree seem to act as
superparticipants by reinforcing the ideas of the group they
belong. Also, most of them only retweet other users inside
the same module (thus, with the same political view). There
were only two exceptions, @kitri1 (April 26th) and
@marcossedassari (May 18th), who retweeted nodes from
both groups. This suggest an action as superparticipant
which differs from those founded by Graham and Wright [4].
The most similar category is superposter, based on their high
active, but the superposters helped users to have a larger
selection of opinions in discussion forums, while the users
with higher outdegree on Twitter act by reinforcing like-
minded messages. We believe these superparticipants with
high outdegree on Twitter could be called as “activists”,
because of their strong alignment with one ideological
position and their effort to reproduce and spread like-minded
tweets.
Further, most of the nodes with higher outdegree seem to
retweet those nodes with higher indegree inside their groups,
reinforcing the idea that while the nodes with higher
indegree are opinion leaders and informational influencers,
the nodes with higher outdegree are responsible for the
spread of the ideas inside each module by their activist
behavior.
On April 26th, for example, the node with highest
outdegree, @beijopai, retweeted all the nodes with higher
indegree in the pro-impeachment group. All other
superparticipants we analyzed also retweeted at least one of
the users with higher indegree inside their groups; the only
two exceptions are @isaceressuelo and @roy_crabbs (both
in April 26th) that retweeted only users from the same
module, but did not retweet those with higher indegree.
These data suggest that the highly active users are
especially responsible for the high indegree of some opinion
leaders and informational influencers we found in our
networks. By reproducing these users’ messages, the
activists make the opinion leaders discourse more visible
inside their group and also give more visibility for the news
they believe help their position.
Our data also suggests that higher outdegree nodes
reinforce the polarized network structure we found in our
analysis. By mostly reproducing like-minded messages
during key political events, they help to building the echo
chambers [22]. We also may consider the activists perform
Influencers in Polarized Political Networks on Twitter SMSociety, July 2018, Copenhagen, Denmark
a selectivity process [7] in which they shared only certain
tweets with a certain political view, which may be one of the
causes of the fragmented structure of political conversation.
Thus, the polarized crowds [23] may also be increased by the
action of the activists in political contexts.
4.3 Discussion
The fragmented structure of the networks we found in our
data might have consequences in the online public sphere.
Further, it might also be affected by some of the influencers
actions (especially the activists) to increase the polarization.
The polarization creates multiple disconnected arenas,
which preclude the communication network among the
various groups necessary to the existence of the public
sphere described by Habermas [9]. This means that the
bridges, the connections among groups, important for the
public sphere and social media relationship [14, 15] were not
found. The contact with others’ views plays an important
role in public sphere [19], but is limited by the isolation
inside the groups. Further, when some users increase the
fragmentation, their action might be a threat to the public
sphere by limiting its democratic function [18].
The characteristics of bubbles and echo chambers cause
a collapse of contexts and might also hinder the individuals
inside the isolated groups from having a full understanding
of the political discussion [17, 22]. Further, individuals
isolated inside a like-minded group tend to not understand
other views and even to develop extremist behavior [22].
Even the power of the influencers tends to be limited.
While they might be strong inside their like-minded groups,
their opinions (in the case of opinion leaders) and
information (in the case of informational influencers) are not
considered in the module they are not a part of. Meanwhile,
the activists limit the content they share by reproducing only
messages with which they agree, creating a selectivity
process [7]. Doing so, they only give visibility to those inside
their module. If the activists shared different points of view
they could become the connections among microspheres
[15], but this did not occur in the networks we analyzed.
5 CONCLUSIONS
In this paper, we have analyzed three political networks to
identify opinion leaders (the users with higher indegree) [1,
3, 5, 6, 9] and superparticipants (the users with higher
outdegree) [4]. We also observed how these influencers act
in the networks and how they helped build the polarized
networks [23] of the echo chambers [17, 22].
We identified in each dataset (1) highly polarized
networks, with the presence of two groups, one pro-
impeachment and the other anti-impeachment, each one
tightly connected. We found (2) three types of influencers:
the “opinion leaders” and the “informational influencers” are
those with higher indegree, while “activists” are those with
higher outdegree. Based on their actions, (3) each of these
types of influencers have a specific way to influence the
debate based on the polarized networks we found. The
opinion leaders are those that assume a specific position and
have their ideas spread among users inside the like-minded
module. The informational influencers are drawn into one
group once its members identify the influencers messages as
favorable for the cluster position. The activists are those that
reproduce the ideas of other like-minded users, especially
the opinion leaders, and in doing so increase the polarized
structure. As for the influencers characteristics, (4) their
position is also related to the polarization. The opinion
leaders reinforce the group position, informational
influencers are drawn into it and activists increase the
fragmentation. Further, (5) we found that the influencers
consistently appear in the same group over time.
Our findings might also indicate that (a) the action of
users (specially the activists) inside the polarized groups
helps the formation of the fragmented network, since they
reinforce like-minded ideas by only reproducing messages
from other users inside the module they are and; (b) it
indicates a threat to the formation of the public sphere on
social media due to the high fragmentation and the absence
of disagreement in the conversations.
We recognize there some limitations present in this
study, such as in only analyzing three specific moments
during the process. Future studies can shed light on different
political conversations to see if the phenomena observed in
this particular case also takes place in other scenarios. Also,
we observed the influencers based on only three metrics. The
analysis of other contexts and based on different metrics
might reveal new categories of influencers. Nevertheless, the
patterns found in the three analyzed networks suggests that
the influencers in political conversations on Twitter tend to
act as one of the three categories we suggested: the opinion
leaders, the informational influencers and the activists.
ACKNOWLEDGMENTS
The authors would like to thank their institutions and CNPq
(projects 305189/2016-6 and 400228/2016-5) for the support which
made this research possible.
REFERENCES
[1]
Meeyoung Cha, Hamed F. Haddadi, Fabricio Benevenuto, Krishna
P. Gummadi. 2010. Measuring user influence on twitter: The
million follower fallacy. In Proceedings of the Fourth
International AAAI Conference on Weblogs and Social Media.
Washington: AAAI, p. 10-17.
[2]
Duncan J. Watts. 2011. Everything Is Obvious Once You Know the
Answer: How Common Sense Fails Us. Crown Business, New
York.
SMSociety, July 2018, Copenhagen, Denmark
F. B. Soares, R. Recuero, G. Zago
[3]
Eytan Bakshy, Jake Hofman, Winter Mason, Duncan J. Watts.
2011. Everyone’s an Influencer: Quantifying influence on Twitter.
In Proceedings of the fourth ACM international conference on Web
search and data mining (WSDM, 2011). ACM Press, Hong Kong,
China, 65-74.
[4]
Todd Graham and Scott Wright. 2013. Discursive equality and
everyday talk online: the impact of “superparticipants”. Journal of
Computer-Mediated Communication 19,3( May 2013), 625-642.
http://onlinelibrary.wiley.com/doi/10.1111/jcc4.12016/full
[5]
Elizabeth Dubois and Devin Gaffney. 2014. The multiple facets of
influence: identifying political influentials and opinion leaders on
Twitter. American Behavioral Scientist, 58, 10 (April 2014), 1260-
1277.
[6]
Paul F. Lazarsfeld, Bernard Berelson and Hazel Gaudet. 1968. The
people’s choice: How the voter makes up his mind in a presidential
campaign. Columbia University Press, New York.
[7]
Bernard R. Berelson and Paul F. Lazarsfeld, William N. McPhee.
1954. Voting: A Study of Opinion Formation in a Presidential
Campaign. University of Chicago Press, Chicago.
[8]
Jürgen Habermas. 1991. The Structural Transformation of the
Public Sphere. The MIT Press, Cambridge.
[9]
Jürgen Habermas. 1996. Between Facts and Norms: Contributions
to a Discourse Theory of Law and Democracy. The MIT Press,
Cambridge.
[10]
Nan Lin. 2001. Social capital. A theory of social structure and
action, Cambridge University Press, Cambridge.
[11]
James S. Coleman. 1988. Social Capital and the Creation of
Human Capital. American Journal of Sociology, n. 94, S95-S120.
[12]
Nicole B. Ellison, Charles Steinfield and Cliff Lampe. 2007. The
benefits of Facebook "friends:" Social capital and college students'
use of online social network sites. Journal of Computer-Mediated
Communication, 12(4), article 1.
http://jcmc.indiana.edu/vol12/issue4/ellison.html.
[13]
Raquel Recuero. 2009. Redes Sociais na Internet. Sulina, Porto
Alegre, Brazil
[14]
Marco T. Bastos. 2012. Public opinion revisited: The propagation
of opinions in digital networks. Journal of Arab & Muslim Media
Research. 4, 2-3 (March 2012), 185-201. From
http://www.intellectbooks.co.uk/journals/view-Article,id=12567.
[15]
Axel Bruns and Tim Highfield. 2016. Is Habermas on Twitter?
Social Media and the Public Sphere. In: Axel Bruns, Gunn Enli,
Eli Skogero, Anders O. Larsson and Christian Christensen (Ed.).
The Routledge Companion to Social Media and Politics.
Routledge, New York (p.56-73).
[16]
Christian Fuchs. 2015Mídias sociais e a esfera pública. Revista
Contracampo, 34, 3, dez/2015-mar/2016. Contracampo (p. 5-80).
[17]
Eli Pariser. 2011. The Filter Bubble. The Penguin Press, New
York.
[18]
Zizi Papacharissi. 2009. The Virtual Sphere 2.0: The Internet, the
Public Sphere and beyond. In: Andrew Chadwick, Philip N.
Howard (eds). Routledge Handbook of Internet Politics.
Routledge, London and New York (p. 230-245).
[19]
Magdalena E. Wojcieszak and Diana C. Mutz. 2009. Online groups
and political discourse: do online discussion spaces facilitate
exposure to political disagreement? In: Journal of communication,
59.1 (March 2009), 40-56.
[20]
Alan Mislove, Bimal Viswanath, Krishna P. Gummadi and Peter
Druschel. 2010. You are who you know. In Proceedings of the
third ACM international conference, ACM Press, New York, USA.
[21]
Kerstin Bischoff. 2012. We love rock “n” roll: analyzing and
predicting friendship links in Last.fm. In Proceedings of the 3rd
Annual ACM Web Science Conference. WebSci '12, Evanston,
USA.
[22]
Cass Sunstein. 2001. Echo Chambers. Princeton University Press,
Princeton.
[23]
Marc Smith, Lee Rainie, Itai Himelboim and Ben Shneiderman.
2014. Mapping Twitter Topic Networks: From Polarized Crowds
to Community Clusters. Pew Research Center, Washington, USA.
[24]
Raquel Recuero, Gabriela Zago and Felipe B. Soares. 2017. Mídia
Social e Filtros-Bolha nas Conversações Políticas no Twitter. In
Proceedings of XXVI Encontro Anual da Compós. Compós, São
Paulo. Brazil.
[25]
Marcelo Alves. 2017. Mobilização da Militância: redes de
camapanha na eleição do Rio de Janeiro em 2016. In Proceedings
of Compolítica 7, Compolítica, Porto Alegre, Brazil.
[26]
Raquel Recuero, Gabriela Zago and Felipe B. Soares. 2017.
Circulação jornalística no Twitter: A cobertura do impeachment
de Dilma Rousseff. In Proceedings of 15º Encontro Nacional de
Pesquisadores em Jornalismo, SBPJor, São Paulo, Brazil.
[27]
Stanley Wasserman and Katherine Faust. 1994. Social Network
Analysis. Cambridge University Press, Cambridge.
[28]
Alain Degenne and Michel Forse. 1999. Introducing Social
Networks. SAGE Publications, London.
[29]
Mark Newman. 2006. Modularity and community structure in
networks. Proc Natl Acad Sci U S A 103 (June 2006), 8577–8582.
[30]
Russell Brandom. 2017. How to spot a Twitter bot. The Verge.
https://www.theverge.com/2017/8/13/16125852/identify-twitter-
bot-botometer-spambot-program.
