Asymmetric polarization

Asymmetric Polarization on Twitter and the 2018
Brazilian Presidential Elections
Felipe Bonow Soares
Federal University of Rio Grande do Sul
Faculdade de Biblioteconomia e
Comunicação
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
Porto Alegre, RS, Brazil
raquelrecuero@gmail.com
Gabriela Zago
Midiars
Pelotas, RS, Brazil
gabrielaz@gmail.com
ABSTRACT
This study aims to understand the dynamics of polarization
and how news outlets influenced political discussions on
Twitter during the 2018 Brazilian presidential campaign.
We analyze four datasets using social network analysis.
Our main finding is the identification of an asymmetric
polarization.
CCS CONCEPTS
• Human-centered computing → Collaborative and
social computing → Collaborative and social computing
theory, concepts and paradigms → Social networks
KEYWORDS
Asymmetric polarization, echo chambers, journalism,
political polarization, social network analysis
ACM Reference format:
Felipe B. Soares, Raquel Recuero and Gabriela Zago. 2019.
Asymmetric Polarization on Twitter and the 2018 Brazilian
Presidential Elections. In Proceedings of the International
Conference on Social Media & Society, Toronto, Canada
(SMSociety). DOI: 10.1145/3328529.3328546
1 INTRODUCTION
During the 2018 Presidential Elections, Brazil faced one of
the most controversial political periods in its history. The
far-right politician Jair Bolsonaro, from the Liberal Social
Party (PSL), was elected with the majority of votes.
Permission to make digital or hard copies of all or part 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 components of this work
owned by others than ACM must be honored. Abstracting with credit is
permitted. To copy otherwise, or republish, to post on servers or to redistribute
to lists, requires prior specific permission and/or a fee. Request permissions
from Permissions@acm.org.
SMSociety '19, July 19–21, 2019, Toronto, ON, Canada
© 2019 Association for Computing Machinery.
ACM ISBN 978-1-4503-6651-9/19/07…$15.00
https://doi.org/10.1145/3328529.3328546
Bolsonaro is known for his controversial positions,
especially regarding women, homosexuality and human
rights 1 . Also, Bolsonaro’s campaign was accused of
disseminating false information during the run2 through
social media channels.
Brazilian political polarization has grown since Dilma
Rousseff’s impeachment in 2016 [24], ending 16 years of
Workers’ Party (PT) government. Rousseff’s impeachment
marked the ascension of new conservative groups in the
country, particularly the far-right, which strongly
supported the impeachment [17]. This also increased an
anti-left movement among Brazilians, particularly against
the Workers’ Party. Further, in 2018, the former-president
Lula da Silva (2003-2010), one of the most popular leaders
of the left, was arrested following a conviction of
corruption, fueling the polarization between left and right-
wings.
Lula da Silva was the Workers’ Party candidate for the
2018 presidential election and the leader in the polls.
However, due to his conviction, he was not allowed to run,
and his vice-president candidate, Fernando Haddad,
stepped up. Bolsonaro’s campaign used the anti-Workers’
Party sentiment as a main strategy in the election, which
reinforced the political polarization. Bolsonaro received
46% of the votes in the first round and Haddad received
29%; then both disputed a runoff, won by Bolsonaro with
55% of the votes.
After his election, many international news outlets
began referring to Bolsonaro as Tropical Trump or Trump
of the Tropics3. The main reason for the comparison is their
1 See https://www.theguardian.com/world/2018/sep/06/jair-bolsonaro-
brazil-tropical-trump-who-hankers-for-days-of-dictatorship.
2 See https://www.theguardian.com/world/2018/oct/10/brazil-fake-news-
presidential-election-whatsapp-facebook.
3 See https://www.telegraph.co.uk/news/2018/10/29/brazils-president-
elect-jair-bolsonaro-really-trump-tropics/ and
https://www.theguardian.com/world/2018/apr/19/jair-bolsonaro-brazil-
presidential-candidate-trump-parallels.
67

SMSociety, July 2019, Toronto, Canada F. B. Soares, R. Recuero, G. Zago
controversial opinions, but they share other characteristics
as well. One of them is that both presidents frequently
criticize and harass mainstream media and journalists who
are not sympathetic to them and give preference to right-
wing partisan media and social media when giving
interviews and announcing political decisions4.
Benkler, Faris and Roberts [4] analyzed the media diet
during elections and the first year of Trump’s government
and found no symmetry between Trump supporters and the
rest of the Americans on social media. Partisan outlets were
more central to the right-wing media ecosystem, while the
left-wing was more diverse. Thus, while left-wing network
had more access to diverse information, right-wing was
more centralized on hyperpartisan information, creating a
polarization between right-wing media and the rest of the
media. Our hypothesis is that a similar phenomenon
occurred during Brazilian elections.
In this paper, we analyze four political conversations in
key moments of the election run. Our data was extracted
from the Twitter API. For this discussion, we will use social
network analysis (SNA) to understand the structure of the
conversations and thus, discuss the role of the media outlets
and the most important nodes for the discussion. Our
research questions are: (1) which dynamic of polarization
characterized the political discussions on Twitter during
Brazilian presidential elections? And (2) which outlets
influenced the political conversation in this period?
2 POLITICAL POLARIZATION AND
ASYMMETRIC POLARIZATION
2.1 Political Polarization and Echo Chambers
In the past few years research has showed evidences of the
presence of political polarization in online conversations,
[3, 4, 11, 12, 23] including in the Brazilian context [1, 22,
24]. In conversations on social media, the polarization
normally creates a network structure called “polarized
crowds” [12, 23]. The polarized crowds are networks
characterized by the presence of two opposing groups with
a high density and with nodes tightly connected to each
other.
Polarized groups are often connected to the rise of echo
chambers [26]. Echo chambers happen when homophilic
groups isolate themselves from other groups with different
ideas, actively limiting the circulation of diverse content. In
echo chambers, individuals only get engaged in discussions
that reinforce their own points of view and become less
receptive to different ideas. In fact, when exposure to
opposing occurs, it serves to further increase the isolation
of echo chambers [2].
Some authors state that the algorithms for social
filtering in social network sites create or strongly increase
4 See https://www1.folha.uol.com.br/internacional/en/world/2018/12/the-
tropical-trump-bolsonaro-follows-closely-the-us-presidents-style.shtml.
the presence of echo chambers [18, 27]. Although these
algorithms influence the creation of a more comfortable
experience for users – and by doing so, they reinforce their
current points of view – social filtering is not the main
culprit in the existence of the phenomenon [4, 24].
Echo chambers are so prevalent that they have been
identified even in contexts where social media was not
relevant [26]. However, in online conversations, users are
often highly active in the formation of echo chambers [24],
engaging themselves in sharing and posting content which
they agree with. Because some of these users play different
roles in the social network, their actions may influence the
whole information diffusion among the group.
2.2 Influencers and Motivated Reasoning
In general, influencers are political leaders or experts in a
particular area and they have the ability or authority to
convince individuals to change or accept an idea or
behavior [6, 9, 14, 24]. Previous research identified three
types of influencers in polarized groups in political
conversations on Twitter [24]: opinion leaders,
informational influencers and activists. While the first two
categories are connected to users with a higher visibility in
the network, activists are highly active users. These
influencers manage to better influence the network by
acting together. The opinion leaders shape the views of the
other users, by being strongly replicated by the activists –
which, in turn, only share like-minded content, reinforcing
the homophily of the group. The informational influencers
are media outlets frequently dragged inside the groups
based on the users’ perception of their messages [1, 21].
The structure of how influencers act in political
discussions is relevant to this research because they
actively increase the gap between echo chambers. By doing
so, they provide a context of support for biased information
when it is favorable for the group’s position [26]. This
behavior could be associated with motivated reasoning.
Motivated reasoning studies contend that individuals
may have their reasoning affected by motivation [13].
Thus, one might reason through a biased cognitive process
to evaluate evidence and form impressions or beliefs based
on what one considers most appropriate, yielding a desired
conclusion. In this process, individuals tend to maintain an
illusion of objectivity and try to logically justify the desired
conclusion. Even when one seeks for confirmation of a
desired conclusion, within an echo chamber the like-
minded opinion leaders and activists tend to reinforce the
biased conclusion if it is favorable for the group narrative.
Soon & Goh [25] explain that this movement often creates
a perception of “false consensus” among the group. Also,
the messages of informational influencers might be
evaluated based on the biased beliefs of the echo chamber,
68

Asymmetric Polarization on Twitter and the 2018 Brazilian Presidential Elections SMSociety, July 2019, Toronto, Canada
when deciding which media outlets are dragged or not to a
group. This process is associated with the group’s media
diet and it might be also associated with the idea of
asymmetric polarization, which we will further discuss.
2.3 Asymmetric Polarization
The visibility given to media outlets inside polarized groups
may be linked to the group’s beliefs and to the presence of
echo chambers. A biased diet of information may also be an
impulse for motivated reasoning, especially when based on
partisan media. Further, these characteristics of echo
chambers with biased reasoning are very likely to create a
perfect environment for disinformation, with intentional
manipulation of information for political purposes [4].
In this study, we consider partisan media distinct from
mainstream media and alternative media. Benkler, Faris &
Roberts [4] define partisan media as those that were clearly
engaged in political campaigns and do not necessarily
follow norms of professional journalism [4]. We adopt this
concept of partisan media in our analysis. We consider as
mainstream media those mass news outlets with more
tradition in journalism. We consider as alternative media
the smaller outlets that provide alternative perspectives.
While partisan media helps to spread disinformation [4],
mainstream and alternative media do not engage in that.
Benkler, Faris & Roberts [4] analyzed the media diet
ecosystem on social media in United States during
presidential election in 2016 and Trump’s first year in 2017.
During this period, they identified two different groups of
actors, one more connected to the right, and the other, to
the left. The left-wing group frequently gave visibility to
center and left mainstream media. On the other hand, the
right-wing group emphasized almost only right-oriented
media, further creating a radicalization pattern within the
group, where more and more far-right biased information
was shared. Therefore, there was no left-right polarization,
but instead a division between the right and the rest of the
media. They called this phenomenon “asymmetric
polarization”.
In the right-wing group, many media outlets
emphasized partisan content over news disseminated by
mainstream media. Further, in some cases these partisan
media even stated that other media outlets that
contradicted and disproved these biased news pieces were
not trustworthy. The authors called this dynamic
“propaganda feedback loop”, because it creates an
environment where media, politicians, activists, and other
participants create a self-reinforcing feedback loop [4]. This
also may increase practices such as firehosing, when real
information perceived as bad for a candidate is fought
through the massive spread of false information [19].
5 https://library.gwu.edu/scholarly-technology-group/social-feed-
manager.
This phenomenon can have negative effects on
democracy and political participation, as polarization and
disinformation increases [8, 15, 20]. Thus, understanding
the political conversations and media diet of different
groups is key to discuss social media effects on democracy.
In this work we will analyze the scenario of political
conversations on Twitter and try to identify patterns of
polarization. Our goal is to understand (1) which
polarization dynamic political discussions on Twitter
during Brazilian presidential elections followed and (2)
which outlets influenced the political conversation in this
period. We will also discuss how media diet might have
influenced Brazilian elections and how it may affect
Brazilian democracy.
3 METHODS
3.1 Data Collection
We used Social Feed Manager 5 to collect tweets that
contained the keywords "Bolsonaro" or “Haddad” (the two
front runners) during September and October of 2018 (the
period of Brazilian presidential election). Data collection
was done on a daily basis, once per hour. From this original
dataset, we filtered four datasets we will use to analyze the
conversations during key dates of the campaign per date.
We analyze data from: 1) September 11th (first round –
keyword “Haddad”), when Haddad was selected as Lula da
Silva substitute as Workers’ Party candidate; 2) September
28th (first round – keyword “Bolsonaro”), when Bolsonaro’s
ex-wife was interviewed by Veja (a Brazilian magazine) and
accused him of concealment of property and other crimes;
3) October 27th (second round – Keyword “Bolsonaro”), one
day before the runoff; and 4) also October 27th (second
round – Keyword “Haddad”).
We decided to analyze October 27th because it was a
“neutral” day on the run, while the other two dates were
selected because one of the candidates was involved in
breaking news of the day. By doing so, we are able to
examine two distinct contexts and come to a more fulsome
conclusion. We did not select a day with negative breaking
news about Haddad because there was none during the
campaign – negative news articles about him were false
and not disseminated by mainstream media.
For this work, we intend to focus on the structure of the
networks to expose polarization and influencers and thus,
will use Social Network Analysis (SNA) [7, 28]. SNA is
useful because it allows us to identify groups in the
conversations, and further observe the position of the news
outlets within those groups. As we are looking at political
conversations on Twitter, each node represents an
individual Twitter account and each connection, a mention
or a retweet.
69

We will work with two metrics: modularity and
indegree. Modularity is used to identify groups in social
networks. The higher the modularity, the denser are the
connections within a module and the less dense the
connections are with other groups [5, 16]. In this analysis,
we used modularity to identify the groups in Twitter
political conversations. Indegree is a metric that shows how
many connections a certain node receives in the network
[10]. A connection in our network represents a mention or
a retweet of a node’s tweet. Thus, actors with high indegree
are those with higher visibility in the conversation. Table 1
provides a breakdown of the data collected.
Table 1: Datasets.
Date Keyword Nodes Tweets Modularity
Sep. 11th Haddad 118,903 264,189 0.518
Sep. 28th Bolsonaro 159,991 423,314 0.504
Oct. 27th Bolsonaro 285,800 628,516 0.543
Oct. 27th Haddad 296,946 648,839 0.531
In order to investigate the presence of an asymmetric
polarization, we first identified the different clusters
involved in each dataset and their ideological focus. We
further analyzed news outlets with the top indegree in each
network, and the module they belonged to. Then, we
classified the news outlets according to their political
position (left, right or center) and the type of content they
circulated more (mainstream, alternative or partisan, as
defined in section 2.3).
We decided to analyze 20 news outlets because they are
the most influential news outlets in the network. Also, we
considered the entire network (and not each group
separately) because this strategy would make it easier to
identify asymmetric polarization, if it exists. We further
qualitatively examined the most retweeted tweets from
each outlet, in order to understand the type of information
that circulates in each cluster.
We named the modules “pro Bolsonaro” and “anti
Bolsonaro” because the pro group is highly centered on
him, while to anti group is not necessarily centered on one
politician, even though it has some centrality to Haddad on
some dates. Also, there was a strong anti Bolsonaro
sentiment during Brazilian election and that is reflected in
our datasets.
4 RESULTS AND DISCUSSION
All the analyzed networks have similar structures, with two
main groups tightly connected (Figures 1-4) and a high
modularity (Table 1). This means the conversations led to a
polarized structure similar to polarized crowds [12, 23].
In all graphs, the green group represents the anti
Bolsonaro discourse and the fuchsia module is pro
Bolsonaro. In all the networks, the anti Bolsonaro group
was larger (more users) than the pro Bolsonaro group.
Below we provide a general description of each of the
networks we analyzed.
Figure 1: Haddad, Sep. 11th.
On September 11th Haddad was announced as Lula da
Silva’s substitute as the Workers’ Party candidate after the
Supreme Courte barred Lula. This is the only network
analyzed in which the green module is highly centered on
a pro Haddad sentiment (rather than an anti Bolsonaro).
Lula da Silva, Haddad, Manuela D’Ávila (Haddad's running
mate) and the Workers’ Party are among the users with
higher indegree.
The fuchsia group is centered in a pro Bolsonaro
sentiment. Danilo Gentili, a humorist and television host
who supported Bolsonaro on the run, was the node with
highest indegree. Bolsonaro himself is not central in the
group, but many of his supporters are among the users with
higher indegree.
Figure 2: Bolsonaro, Sep. 28th.
On September 28th, the Brazilian magazine Veja
published a polemic interview with Bolsonaro’s ex-wife. In
this interview, she accused him of concealment of property
and other crimes.
70

The anti Bolsonaro group is especially heterogeneous.
The first round of Brazilian election was in October 7th, so
in September 28th there were still 13 candidates in the
running. Some of them appear among the users in the green
module. Ciro Gomes (a candidate that received 12% of the
votes in the first round) was the one with a higher indegree,
followed by Geraldo Alckmin (4.76% of the votes) and
Guilherme Boulos (0.58% of the votes). Ciro Gomes was
associated with a center-left political position, while
Geraldo Alckmin was associated with a center-right
position and Guilherme Boulos was associated with a left
position. Manuela D’Ávila, also appeared in the group,
although Haddad himself had little visibility. This further
establishes that the green module does not represent a “pro
Haddad” position, but rather an “anti Bolsonaro” position.
The fuchsia module gave a high visibility for the
musician Roger Moreira, who supported Bolsonaro on the
run. Also, Bolsonaro himself and his three politician sons,
Eduardo, Flavio, and Carlos, are some of the main users in
the group. This shows a high association of the module
with Bolsonaro.
Figure 3: Bolsonaro, Oct. 27th.
We decided to include October 27th because it was a
more neutral day during the campaigns, with no candidate-
specific news – the runoff was on October 28th.
Nevertheless, we found a polarized network very similar to
the others.
Joaquim Barbosa, a former Brazilian Chief Justice who
sent to jail some of the Workers’ Party leaders for
corruption, declared his vote for Haddad and criticized
Bolsonaro’s anti-democratic views. This made Barbosa the
user with the highest visibility in the anti Bolsonaro
module (green), followed by Haddad himself. Felipe Neto, a
popular Brazilian YouTube star, who also criticized
Bolsonaro and declared his vote for Haddad, had high
centrality as well. The green module had a strong anti
Bolsonaro sentiment, more than a pro Haddad position.
The fuchsia group kept its characteristics compared to
September 28th, with Bolsonaro as the user with a higher
indegree. His son Flavio also received high visibility, among
other Bolsonaro supporters.
Figure 4: Haddad, Oct. 27th.
The network of October 27th collected using “Haddad”
as keyword is very similar to the one with the keyword
“Bolsonaro” (Figure 3). Once again, Joaquim Barbosa is the
node with higher indegree within the anti Bolsonaro group
(green), followed by Haddad and Felipe Neto. It shows that
even when “Haddad” is used as keyword during the runoff
there was a strong anti Bolsonaro sentiment in the green
module. The pro Bolsonaro module (fuchsia) is centered in
Bolsonaro himself as the user with highest indegree,
followed by other politicians who supported him.
The pattern found in all four datasets indicates that the
main polarization during Brazilian presidential elections
was between Bolsonaro and the rest of the candidates. As
seen on September 28th, for example, there was a very
heterogeneous environment in the green module. On the
other hand, the fuchsia module was centered in a pro
Bolsonaro discourse along all four datasets. It is similar to
what have been found by Benkler, Faris and Roberts [4] in
the US scenario.
4.1 Mapping News Outlets
We identified the 20 news outlets with highest indegree in
each of the four networks and qualitatively examined their
accounts. Then, we mapped the module they belong to.
Finally, we categorized then as mainstream media, partisan
media, and alternative media (as defined in section 2.3), and
according to their political position (left, right, or center).
We also identified the most retweeted tweet from each
outlet to further examine how the retweeting behavior is
connected to the content of the tweet.
September 11th was the day Workers’ Party announced
Haddad as Lula da Silva substitute. Among the four
networks we analyzed, this is the least biased media diet in
the pro Bolsonaro group (Table 2). Despite of it, we found
in this group three right-wing news outlets and two of
them were identified as partisan media. Veja, the news
71

outlet with higher indegree in the Pro Bolsonaro group,
historically criticized the Workers’ Party and is identified
as an anti-left-wing magazine. Among the most retweeted
tweets of these news outlets, some of them mentioned polls
favorable to Bolsonaro and unfavorable to Haddad. Veja
tweeted: “Datafolha: On the second round, Bolsonaro is
defeated by everyone, except by Haddad6”; and G1: “Ibope
Poll: Bolsonaro, 26%; Ciro, 11%; Marina, 9%; Alckmin, 9%;
Haddad, 8%”. While these tweets are factual, they point to
Haddad’s disadvantage in the polls. There was also more
poignant negative news about Haddad, such as in this tweet
from O Antagonista: “Lula will set up Haddad’s ministry7”
and from G1 Política: “Marina says Haddad is ‘similar’ to
Dilma and voting for an ‘heir’ can take Brazil to ‘rock
bottom’8 ”. Both tweets make reference to the fact that
Haddad was portrayed both by the Worker’s Party and the
opposition as Lula’s heir. However, while the Worker’s
Party tried to use this as an advantage (because of Lula’s
popularity), the opposition framed Haddad as a puppet.
Among anti Bolsonaro users (Table 3) we found a
majority of mainstream media, although many of them are
left-wing vehicles. This happened because the center
mainstream media focused more on the announcement, for
example: UOL Notícias: “Worker’s Party Leadership
approves Haddad’s name as the new party candidate to the
presidency #UOLintheballots 9 ”. Left-wing outlets also
demonstrated their support to the candidate: Carta Capital:
“Carta Capital, following the tradition of publications from
Europe and USA, supported both Lula and Dilma on their
elections. From now on, we will support Fernando Haddad,
Lula’s and our candidate. Read the editorial by Mino
Carta”10. As mentioned before, this is the only network
where there is a pro Haddad sentiment within the green
group, rather an anti Bolsonaro. It is caused by the political
context (it was still the first round) and the news agenda of
the day (Haddad as the new Workers’ Party candidate).
Table 2: “Haddad”, Sep. 11th, Pro Bolsonaro
News outlets Type Position Indegree
Veja Mainstream Right 997
O Antagonista Partisan Right 620
G1 Mainstream Center 545
Estadão Mainstream Center 490
G1 Política Mainstream Center 417
UOL Mainstream Center 302
Conexão Política Partisan Right 266
6 Translated from Portuguese: “No Segundo turno, Bolsonaro é derrotado
por todos, menos por Haddad”.
7 Translated from Portuguese: “Lula vai montar o ministério de Haddad”.
8 Translated from Portuguese: “Marina diz que Haddad é ‘semelhante’ a
Dilma e que votar em ‘indicado’ pode levar Brasil ao ‘poço sem fundo’”.
9 Translated from Portuguese: “Executiva do PT aprova nome de Haddad
como novo candidato à Presidência #UOLnasUrnas”.
10 Translated from Portuguese: “CartaCapital, conforme tradição de
publicações da Europa e EUA, apoiou as duas candidaturas de Lula e de
Table 3: “Haddad”, Sep. 11th, Anti Bolsonaro
Folha de S. Paulo Mainstream Center 1,835
Carta Capital Mainstream Left 1,412
Brasil 247 Partisan Left 1,353
Diário do Centro do Mundo Partisan Left 932
UOL Notícias Mainstream Center 835
Jornalistas Livres Alternative Left 737
El País Brasil Mainstream Left 710
Brasil de Fato Alternative Left 589
Revista Fórum Mainstream Left 560
O Globo Mainstream Center 484
Folha Poder Mainstream Center 422
Catraca Livre Alternative Left 386
The Intercept Brasil Alternative Left 298
On the September 28th dataset, Veja was the news outlet
with the highests indegree (it was also the node with the
highest indegree considering all types of users). This was
expected, considering that Veja interviewed Bolsonaro’s
ex-wife, where she accused him of several crimes (the main
news event of the date, and the reason we selected this date
to be part of our research). Veja appeared in the anti
Bolsonaro group (Table 5) along with many other
mainstream news outlets, most of them with no clear
political position (classified as center). Most of them
reverberate Veja’s breaking news. Veja tweeted:
“EXCLUSIVE: In a legal process of more than 500 pages to
which Veja had access, Bolsonaro’s ex-wife accuses him of
stealing a vault, hiding his assets, receiving non-declared
payments, and acting with ‘unlimited aggressiveness’11”;
while Folha de S. Paulo tweeted: “Ex-wife accuses
Bolsonaro of stealing vault, and aggressiveness12”. In both
cases, the cluster gives visibility to the scandal, and how
badly it portrayed Bolsonaro.
Among the 20 top indegree news outlets on September
28th, only four are within the pro Bolsonaro group (Table
4). All of them are partisan media that supported him on
the run. This shows an extreme biased media diet among
Bolsonaro’s supporters. Also, in their most retweeted
tweets these partisan media criticized mainstream media
and said the breaking news about Bolsonaro’s ex-wife was
not trustworthy, similarly to what have been found by
Benkler, Faris and Roberts [4]. Politz Oficial tweeted: “Good
morning Brazilians! Today, at dawn, we published the
information exclusively: The Abril Publishing was
responsible for reopening a family lawsuit involving
Dilma. A partir desta edição apoiamos Fernando Haddad, candidato de
Lula e nosso"| Leia editorial de Mino Carta”.
11 Translated from Portuguese: “EXCLUSIVO: Em um processo de mais de
500 páginas, às quais VEJA teve acesso, ex-mulher acusa Bolsonaro de
furtar cofre, ocultar patrimônio, receber pagamentos não declarados e agir
com ‘desmedida agressividade’”.
12 Translated from Portuguese: “Ex-mulher acusou Bolsonaro de furto de
cofre e agressividade”.
72

Bolsonaro and his ex-wife. Bolsonaro WASN’T THE
DEFENDANT. He was the AUTHOR of the lawsuit against
her. Share!13”; and Conexão Política tweeted: “BREAKING
NEWS: VEJA X Bolsonaro: the sensationalism behind a
lawsuit about alimony and child custody14”.
Table 4: “Bolsonaro”, Sep. 28th, Pro Bolsonaro
Renova Mídia Partisan Right 3,437
Politz Oficial Partisan Right 2,300
Conexão Política Partisan Right 2,291
O Antagonista Partisan Right 1,689
Table 5: “Bolsonaro”, Sep. 28th, Anti Bolsonaro
Veja Mainstream Right 10,915
Mídia NINJA Alternative Left 1,954
Revista Época Mainstream Center 1,909
Estadão Mainstream Center 1,765
Revista Exame Mainstream Center 1,633
G1 Mainstream Center 1,069
Diário do Centro do Mundo Partisan Left 911
UOL Notícias Mainstream Center 735
Brasil 247 Partisan Left 689
Carta Capital Mainstream Left 598
Jota.Info Mainstream Center 472
InfoMoney Mainstream Center 451
O Globo Mainstream Center 402
BuzzFeed News Brasil Mainstream Center 373
October 27th was considered a “neutral” day – there was
no breaking news about any candidate. Nevertheless, the
Pro Bolsonaro module in the network with “Bolsonaro” as
keyword (Table 6) had less news outlets among the top 20
indegree and is once again centered on right-wing partisan
media. This means that the biased media diet was not
caused by negative news about Bolsonaro, rather, it
represented ordinary behavior throughout the run.
Mainstream outlets only appear among those with top
indegree because their messages are positive for Bolsonaro.
For example, Estadão tweeted: “’Bolsonaro will set the tone
for a pacified country’, says Doria15”16; and Veja tweeted:
13 Translated from Portuguese: “Bom dia brasileiros! Hoje na madrugada
publicamos a informação com exclusividade: A Editora Abril foi a
responsável pelo desarquivamento do processo familiar de Bolsonaro e sua
ex-esposa. Bolsonaro NÃO ERA RÉU no processo. Ele foi o AUTOR da
ação contra ela. Divulguem!”.
14 Translated from Portuguese: ÚLTIMAS: VEJA X Bolsonaro: o
sensacionalismo por trás de um processo de pensão alimentícia e guarda”
15 Doria is a right-wing politician and supported Bolsonaro candidature.
16 Translated from Portuguese: Estadão: “‘Bolsonaro é quem dará o tom de
um país pacificado’, diz Doria”.
17 Translated from Portuguese: “Euforia com Bolsonaro leva dólar a menor
cotação desde maio”.
“Euphoria with Bolsonaro takes US dollar to the lowest
price since May17”.
The anti Bolsonaro group (Table 7) shows a wider
variety of news outlets, including many mainstream media.
Even The Economist appears among anti Bolsonaro users
because the British magazine criticized the then-candidate.
The Economist tweeted: “If Jair Bolsonaro wins Brazil's
election, the survival of democracy in Latin America's
largest country could be put at risk”. There were also some
tweets about anti Bolsonaro demonstrations, such as in
tweets from Folha de S. Paulo: “Roger Waters protests
against Bolsonaro within the legal time18”19; and Mídia
NINJA: “Why not Bolsonaro? See the words of
@emicida20!”21.
Table 6: “Bolsonaro”, Oct. 27th, Pro Bolsonaro
Revista Exame Mainstream Center 1,741
G1 Mainstream Center 920
Table 7: “Bolsonaro”, Oct. 27th, Anti Bolsonaro
UOL Notícias Mainstream Center 2,288
O Globo Mainstream Center 2,115
Jornalistas Livres Alternative Left 1,514
Diário do Centro do Mundo Partisan Left 1,174
The Economist Mainstream Right 1,168
Revista Fórum Mainstream Left 814
Congresso em Foco Mainstream Center 753
Carta Capital Mainstream Left 749
El País Brasil Mainstream Left 726
Both networks from October 27th had similar results. On
the one using “Haddad” as keyword, the pro Bolsonaro
module (Table 8) gave more visibility to Veja, but three
partisan media are among top 5 indegree news outlets.
18 The musician Roger Waters criticized Bolsonaro in his shows in Brazil,
calling the politician a neo-fascist. See:
https://www.theguardian.com/music/2018/oct/10/roger-waters-jair-
bolsonaro-brazil-concert.
19 Translated from Portuguese: “Roger Waters se manifesta contra
Bolsonaro dentro do horário permitido por lei”
20 Brazilian musician.
21 Translated from Portuguese: “Por que Bolsonaro não? Com a palavra,
@emicida!”.
73

Once again, there is a majority of right-wing vehicles. Veja
and the partisan media O Antagonista mentioned a
Brazilian musician that asked Haddad to delete a video of
supporting him that happened to use one of her songs. Veja
tweeted: “Paula Toller asks to delete videos of support to
Haddad with Kid Abelha 22 ”. O Antagonista tweeted:
“Electoral Supreme Court orders shut down of supporter’s
video for Haddad that uses music from Paula Toller23”.
There were other negative news for Haddad, such as in the
O Globo Política tweet: “Regional Electoral Court opens
investigation about irregular support from Paraiba
government to @Haddad_Fernando24”; and Renova Mídia
tweeted: “The Brazilian Bar Association published a note
with three other entities denying Haddad’s claim25 and
explaining they didn’t officially support any of the
candidates26 ”. Once again, mainstream media only had
centrality among pro Bolsonaro users because their news
supported the pro Bolsonaro narrative.
Within the anti Bolsonaro module (Table 9), there is a
majority of mainstream media and a balance between
center and left vehicles. Among the top five, four of them
are mainstream media and three do not express a political
position – including Folha de S. Paulo and O Globo, two
traditional news outlets in Brazil. Many news outlets
reverberate Joaquim Barbosa’s (the node with highest
indegree in the network) message. Folha de S. Paulo
tweeted: “Joaquim Barbosa, ex-president of the Supreme
Court, declares vote for Haddad27”. Some news outlets also
mention other Brazilian public figures, such as the
journalist Marcelo Tas and the musician Paulinho da Viola,
as in Brasil 247’s tweet: “Workers’ Party’s critic, Marcelo
Tas declares his vote for Haddad”28 and Revista Fórum’s:
“Paulinho da Viola declares vote for Haddad: ‘Black lives
are not measured in arrobas29, they are measured in song
beats, verses’30”.
Table 8: “Haddad”, Oct. 27th, Pro Bolsonaro
O Globo Política Mainstream Center 1,930
G1 Mainstream Center 1,054
Revista Istoé Mainstream Center 555
22 Translated from Portuguese: “Paula Toller pede exclusão de vídeos de
apoio a Haddad com Kid Abelha”.
23 Translated from Portuguese: “TRE manda retirar vídeo de apoio a
Haddad que usa música de Paula Toller”.
24 Translated from Portuguese: “TSE abre investigação sobre suposto apoio
irregular do governo da Paraíba a Haddad”.
25 Workers’ Party mentioned on their campaign website that the Brazilian
Bar Association supported Haddad on the run.
26 Translated from Portuguese: “Ordem dos Advogados do Brasil (OAB)
divulgou nota em conjunto com outras três entidades desmentindo a
Table 9: “Haddad”, Oct. 27th, Anti Bolsonaro
O Globo Mainstream Center 3,843
Revista Fórum Mainstream Left 2,440
UOL Notícias Mainstream Center 2,111
Diário do Centro do Mundo Partisan Left 2,052
Jornalistas Livres Alternative Left 1,627
Carta Capital Mainstream Left 1,148
Revista Exame Mainstream Center 920
Jornal do Brasil Mainstream Center 750
Brasil de Fato Mainstream Left 614
All four datasets follow the same pattern. In all of them,
the anti Bolsonaro group had more news outlets in number
among the top 20 indegree. Also, the same group had
stronger presence of mainstream media, while partisan
media had higher centrality in the pro Bolsonaro group.
The indegree of the media outlets is also considerably
higher in the anti Bolsonaro module – meaning that media
outlets have higher centrality within these groups. These
characteristics suggest that anti Bolsonaro users gave
visibility to a wider variety of media outlets, maintaining
more diverse media diet than those pro Bolsonaro.
4.2 Discussion
In this work we analyzed four Twitter networks associated
with the 2018 Brazilian presidential elections. As we
highlighted above, in all four datasets the “anti Bolsonaro”
group had more news outlets in number among top 20
indegree and also a wider variety of them. While the
visibility in the “pro Bolsonaro” module is given mostly to
partisan media, in the “anti Bolsonaro” group there are
more mainstream media outlets among the nodes with
higher indegree. This means that the information flow is
different within each group.
In three of the four networks we analyzed there is a
majority of right-wing news outlets in the “pro Bolsonaro”
group. Further, in all of them there is also a strong presence
of partisan media. These patterns are especially strong
around September 28th, when there was a negative story
about Bolsonaro. In this case, partisan media tried to
disqualify the negative information by spreading a story
reporting that the magazine that broke it had been paid 600
campanha de Fernando Haddad (PT) e explicando que não manifestaram
apoio a nenhum dos presidenciáveis.”.
27 Translated from Portuguese: “Joaquim Barbosa, ex-presidente do STF,
Joaquim Barbosa declara voto em Haddad”.
28 Translated from Portuguese: “Crítico do PT, Marcelo Tas declara voto
em Haddad”.
29 Weight measure used for merchandise by Portuguese and Spanish
colonies.
30 Translated from Portuguese: "Paulinho da Viola declara voto em
Haddad: ‘Negro não se mede em arroba, se mede em batuques, versos’”.
74

million by the Worker’s Party to vilify Bolsonaro. This
practice is known as “firehose” and happens when partisan
media tries to bury a bad story about a politician through a
torrent of false stories [19]. Often, the false stories circulate
more among their supporters than the real ones, which was
the case here. By doing so, these outlets are able to block
certain content and to maintain a positive narrative about
Bolsonaro among his supporters through the echo chamber
[26]. Further, the partisan media narrative reinforces a
biased comprehension of the facts, also making motivated
reasoning [13] more likely among the pro Bolsonaro users.
Mainstream media only appear within the pro Bolsonaro
group when their narrative was used to validate biased
news from partisan media. This suggests that traditional
media outlets do circulate in echo chambers, but only when
their stories are perceived as acceptable (in this case,
reinforcing the pro Bolsonaro narrative).
Among the different types of influencers [6, 9, 14, 24],
news outlets are normally associated with the idea of
informational influencers [24]. These users often propagate
new content among social media networks. Informational
influencers could be a source of different points of view, but
when clusters have a media diet based in partisan media,
users reject this possibility. This means that partisan media
does not play the proper role of informational influencers,
rather they reinforce biased, like-minded narratives, as we
observed in the pro Bolsonaro networks.
On the other hand, the “anti Bolsonaro” group had a
more heterogeneous media diet, as well as more diversity
of discourses. This is especially clear in the datasets of
September 28th, when there were more candidates in the
presidential run, and in October 27th, when other parties
and politicians were declaring their vote for Haddad. While
the “pro Bolsonaro” group was strongly identified with
him, the “anti Bolsonaro” module was not associated only
with Haddad, even though he had an important role in most
of the datasets we analyzed.
The variety in the anti Bolsonaro media diet indicates a
smaller chance of the formation of echo chambers [26].
Further, the presence of a majority of mainstream media is
unlikely to reinforce biased conclusions, making motivated
reasoning [13] less likely among anti Bolsonaro users. The
presence of international news outlets, such as The
Economist, and Brazilian versions of El País and The
Intercept in the “anti Bolsonaro” module provides further
evidence of their heterogeneous media diet. In the case of
the anti Bolsonaro group, media outlets in fact play out
their role as informational influencers [24], because the
variety of outlets provides heterogeneous content from
different points of view. In these groups, we did not find
fabricated news among tweets with more circulation, as we
did in the others, which is also an evidence of a plurality of
information.
Based on our findings, we infer the existence of an
asymmetric polarization [4] during political conversations
surrounding the Brazilian election. While anti Bolsonaro
users consumed a wider variety of news outlets, pro
Bolsonaro users had a very limited media diet centered in
partisan media. As mentioned, when there was a negative
story about Bolsonaro, this pattern was stronger. Pro
Bolsonaro clusters created what Benkler, Faris and Roberts
[4] call a self-reinforcing feedback loop, the main
characteristic of the disinformation in an asymmetric
polarized group.
As polarization and disinformation increases, this can
have negative effects on democracy and political
participation [8, 15, 20]. The more people consume only
partisan stories, the more they tend to rely on “false
consensus” [25] and justify extreme positions through
motivated reasoning [13]. Considering that partisan media
increase disinformation in order to strength a narrative that
favors their ideology [4], the political participation of
people only exposed to this type of content may be
impaired.
Thus, we can argue that this asymmetric polarization
may have negatively influenced the Brazilian election
results. The biased media diet of pro Bolsonaro users tends
to strengthen radicalization of the far-right and consolidate
biased conclusions among these groups. This means that
pro Bolsonaro users were more likely to consume and trust
biased information in order to maintain the coherence of
their narrative. By doing so, they created an alternative
narrative of the facts and negatively influenced political
conversations during the elections.
5 CONCLUSIONS
In this paper, we analyzed four political networks to
understand the dynamics of polarization during the 2018
Brazilian presidential elections. In our datasets we
identified the presence of two groups. One of them was
highly centered in Bolsonaro, which we called the “pro
Bolsonaro” group. The other one supported Haddad to an
extent, but more than that, shared an anti Bolsonaro
sentiment; and we decided to define this second group as
“anti Bolsonaro” rather than pro Haddad. In each dataset
we identified the top 20 indegree news outlets and mapped
them in the network in order to understand the users’
media diet. We qualitatively analyzed these news outlets
and identified their most retweeted message in each
dataset.
In response to our first research question, we concluded
that political discussions on Twitter during Brazilian
presidential campaign followed a dynamic of an
asymmetric polarization [4]. Our second research question
focused on the types of outlets that influenced the
conversation. The pro Bolsonaro users presented a
radicalized behavior by giving high centrality to right-wing
partisan media. By doing so, they preserved a biased
narrative created to support Bolsonaro. Mainstream media
75

only received visibility among pro Bolsonaro users when
they helped to consolidate this narrative. On the other
hand, anti Bolsonaro users had a more varied media diet. In
all datasets we analyzed, the anti Bolsonaro group had
more news outlets in number among the top 20 indegree.
Anti Bolsonaro users also consumed news from a wider
variety of sources, including Brazilian mainstream media,
many of them with center political views, and even some
international news outlets, such as The Economist.
We recognize there are some limitations in this study,
such as in only analyzing four datasets. Also, on September
28th we analyzed how a negative breaking news about
Bolsonaro reverberated, but on September 11th the breaking
news about Haddad was not negative. We had no clearly
negative breaking news about Haddad during the run
(except some fabricated false news), so we had to select the
discussion and reaction to his nomination as candidate to
analyze.
Future studies might further analyze the discursive
patterns of the anti and pro Bolsonaro groups. Also, future
studies can analyze if the asymmetric polarization pattern
remains the same during Bolsonaro’s first year as president
(in 2019).
REFERENCES
[1] 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.
[2] Christopher A. Bail, Lisa P. Argyle, Taylor W. Brown, John P.
Bumpus, Haohan Chen, M. B. Fallin Hunzaker, Jaemin Lee,
Marcus Mann, Friedolin Merhout and Alexander Volfovsky. 2018.
Exposure to opposing views on social media can increase political
polarization. PNAS, 115, 37, 9216–9221.
[3] Marco Bastos, Dan Mercea and Andrea Baronchelli. 2017. The
Spatial Dimension of Online Echo Chambers. [physics.soc-ph].
arXiv:1709.05233v1. From
https://arxiv.org/ftp/arxiv/papers/1709/1709.05233.pdf.
[4] Yochai Benkler, Robert Faris and Hal Roberts. 2018. Network
Propaganda: Manipulation, disiformation, and radicalization in
american politics. Oxford University Press, New York, NY.
[5] Vincent D. Blondel, Jean-Loup Guillaume, Renaud Lambiotte and
Etienne Lefebvre. 2008. Fast unfolding of communities in large
networks. [physics.soc-ph]. From
http://lanl.arxiv.org/abs/0803.0476.
[6] 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.
[7] Alain Degenne and Michel Forse. 1999. Introducing Social
Networks. SAGE Publications, London.
[8] Hossein Derakhshan and Claire Wardle. 2017. Information
Disorder: Definitions. In Proceedings of Understanding and
Addressing the Disinformation Ecosystem. Annemberg: University
of Pennsylvania, 5-12.
[9] Elizabeth Dubois and Devin Gaffney. 2014. The multiple facets of
influence: identifying political influentials and opinion leaders on
Twitter. American Behavioral Scientist, 58, 10, 1260-1277.
[10] Linton C. Freeman. 1979. Centrality in Social Networks:
Conceptual clarification. Social Networks, 1, 215-239.
[11] Anatoliy Gruzd and Jeffrey Roy. 2014. Investigating Political
Polarization on Twitter: A Canadian Perspective. Policy and
Internet, 6, 1, 28-45.
[12] Itai Himelboim, Marc Smith, Lee Rainie, Ben Schneiderman,
Camila Espina. 2017. Classifying Twitter Topic-Networks Using
Social Network Analysis. Social Media + Society, 3, 1.
[13] Ziva Kunda. 1990. The Case for Motivated Reasoning.
Psychological Bulletin, 108, 3, 480-498.
[14] 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.
[15] Alice Marwick and Rebecca Lewis. 2018. Media Manipulation and
Disinformation Online. Data & Society Research Institute, New
York, NY.
[16] Mark Newman. 2006. Modularity and community structure in
networks. Proc Natl Acad Sci U S A 103 (June 2006), 8577–8582.
[17] Francisco M. Oliveira. 2017. A ascensão do conservadorismo no
Brasil: surgimento e atuação de movimentos sociais
conservadores. Proceedings of ALAS 2017. Montevideo, Uruguay.
From
http://alas2017.easyplanners.info/opc/tl/1445_francisco_mesquit
a_de_oliveira.pdf.
[18] Eli Pariser. 2011. The Filter Bubble. The Penguin Press, New York.
[19] Christopher Paul and Miriam Matthews. 2016. The Russian
"Firehose of Falsehood" Propaganda Model: Why It Might Work
and Options to Counter It. RAND Corporation, Santa Monica,
CA. From https://www.rand.org/pubs/perspectives/PE198.html.
[20] Michael A. Peters. 2018. The information wars, fake news and the
end of globalisation, Educational Philosophy and Theory, 50, 13,
1161-1164, 2018 . DOI: 10.1080/00131857.2017.1417200.
[21] 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.
[22] 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.
[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] Felipe Bonow Soares, Raquel Recuero and Gabriela Zago. 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
[25] Carol Soon and Shawn Goh. 2018. Fake news, false information
and more: Countering human biases. IPS Working Papers, 31.
From https://lkyspp.nus.edu.sg/docs/default-source/ips/ips-
working-paper-31_fake-news-false-information-and-
more_260918.pdf.
[26] Cass Sunstein. 2001. Echo Chambers. Princeton University Press,
Princeton.
[27] Cass Sunstein. 2017. #Republic. Princeton University Press,
Princeton.
[28] Stanley Wasserman and Katherine Faust. 1994. Social Network
Analysis. Cambridge University Press, Cambridge.
76

Asymmetric polarization

More Related Content

What's hot (18)

Similar to Asymmetric polarization (20)

Recently uploaded (20)

Asymmetric polarization