Temporal Analysis of Brazilian Presidential Election on Twitter Based on
Formal Concept Analysis
Daniel Pereira
a
, Julio Neves
b
, Wladmir Brand
˜
ao
c
and Mark Song
d
Universidade Cat
´
olica de Minas Gerais, Computer Science Department, Minas Gerais, Brazil
Keywords:
Topic Evolution, Twitter, Formal Concept Analysis, Social Network Analysis.
Abstract:
Social networks have become an environment where users express their feelings and share news in real-time.
However, analyzing the content produced by users is not a simple task, given the volume of posts. It is
important to comprehend the expressions made by users to gain insights into politicians, public figures, and
news. The state-of-the-art lacks studies that propose how the topics discussed by social network users change
over time. In this context, this work measures how topics discussed on Twitter vary over time. Formal Concept
Analysis was used to measure how these topics were varying, considering the support and confidence metrics.
Our solution was tested on tweets related to the Brazilian presidential election. The results confirm that it
is possible to comprehend what Twitter users were discussing and how these topics changed over time. Our
work is beneficial for politicians seeking to analyze the discussions about them among users. Our analysis of
3,634 tweets revealed several significant patterns, such as the association between political figures and topics
like fake news and election fraud. These findings demonstrate how social media discussions evolve during key
political events, providing insights that can assist political campaigns in real-time.
1 INTRODUCTION
The Internet is no longer just a repository for docu-
ments to be shared, it is now a hybrid space for differ-
ent media and applications that reach a large audience
(Zhang et al., 2012). Some of these applications are
social networks, which allow their users to generate a
large amount of content that exemplifies their impres-
sions and experiences. A specific social network that
stands out for forcing its users to express themselves
concisely is Twitter. On Twitter, users express them-
selves through tweets, which consist of text content
with a maximum length of 280 characters.
The fact that the Tweet is a short textual model
allows users to quickly report what they are experi-
encing at the time the post is posted, unlike a jour-
nalist, for example, who, to generate a story, needs
to ensure its excellence. Since Twitter users report
their experiences without worrying about their writ-
ing or who will read their text, Twitter is probably
the fastest means of disseminating information in the
a
https://orcid.org/0009-0005-2276-2334
b
https://orcid.org/0000-0002-0520-9976
c
https://orcid.org/0000-0002-1523-1616
d
https://orcid.org/0000-0001-5053-5490
world (Cataldi et al., 2010).
With this large amount of information provided, it
is hard to extract knowledge from a group of tweets.
This task is relevant for politicians, for example, to
check the opinions that users are expressing about
them. Therefore, it would be relevant to develop a tool
that is capable of analyzing and extracting knowledge
from a group of tweets.
An alternative to solve this challenge is through
Natural Language Processing (NLP) and Formal Con-
cept Analysis (FCA). The objective of our work is
to use NLP to find recurrent groups of words from
tweets and then analyze how these groups of words
relate to each other. The relation between these terms
is measured using FCA, using the metrics of support
and confidence. Also, how these terms change over
time is another metric analyzed.
While social media platforms like Twitter offer
rich datasets for sentiment analysis, existing methods
often fall short in tracking topic evolution over time.
Traditional NLP techniques primarily focus on static
analysis, which limits their ability to capture the dy-
namic nature of online discussions. This work aims
to address these gaps by employing Formal Concept
Analysis FCA to observe how political topics evolve
during critical events, such as elections.
Pereira, D., Neves, J., Brandão, W. and Song, M.
Temporal Analysis of Brazilian Presidential Election on Twitter Based on Formal Concept Analysis.
DOI: 10.5220/0012869300003825
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 20th International Conference on Web Information Systems and Technologies (WEBIST 2024), pages 167-174
ISBN: 978-989-758-718-4; ISSN: 2184-3252
Proceedings Copyright © 2024 by SCITEPRESS – Science and Technology Publications, Lda.
167

A case study to solve the problem was used. The
case study consists of analyzing tweets that discuss
the Brazilian presidential election, checking which
terms are related to each candidate and how they
evolve over time.
The remainder of the paper is organized as fol-
lows: the background is outlined in Section 2. The
Literature Review is described in Section 3. Section
4 presents the defined Methodology. Results are dis-
cussed in Section 5. The conclusion and further re-
search are in Section 6.
2 BACKGROUND
2.1 Formal Concept Analysis
FCA is a technique based on formalizing the notion
of concept and structuring concepts in a conceptual
hierarchy. FCA relies on lattice theory to structure
formal concepts and enable data analysis. The ca-
pability to hierarchize concepts extracted from data
makes FCA an interesting tool for dependency anal-
ysis. With the increase of social networks and due to
the large amount of data generated by users, the study
and improvement of techniques to extract knowledge
are becoming increasingly justified. Also, it permits
the data analysis through associations and dependen-
cies attributes, and objects, formally described, from
a dataset.
Formally, a formal context is formed by a triple
(G, M, I), where G is a set of objects (rows), M is a set
of attributes (columns) and I is defined as the binary
relationship (incidence relation) between objects and
their attributes where I ⊆ G × M.
Table 1 exemplifies a formal context. In this ex-
ample, objects correspond to tweets, attributes are the
characteristics (terms), and the relationship of inci-
dence represents whether or not the tweet has that
characteristic. An
′
X
′
is present in the table if the
tweet possesses the corresponding characteristic.
Table 1: Formal Context Example.
Lula Bolsonaro Fake News Elections
Tweet 1 X
Tweet 2 X X
Tweet 3 X X X
Tweet 4 X
2.2 Formal Concepts
Let (G, M, I) be a formal context, A ⊆ G a subset
of objects and B ⊆ M a subset of attributes. Formal
concepts are defined by a pair (A, B) where A ⊆ G is
called extension and B ⊆ M is called intention. This
pair must follow the conditions where A = B
′
and
B = A
′
(Ganter and Wille, 1999). The relation is de-
fined by the derivation operator (
′
):
A
′
= { m ∈ M| ∀ g ∈ A, (g, m) ∈ I}
B
′
= { g ∈ G| ∀ m ∈ B, (g, m) ∈ I}
If A ⊆ G, then A
′
is a set of attributes common to
the objects of A. The derivation operator (
′
) can be
reapplied in A
′
resulting in a set of objects again (A
′′
).
Intuitively, A
′′
returns the set of all objects that have
in common the attributes of A
′
; note that A ⊆ A
′′
. The
operator is similarly defined for the attribute set. If
B ⊆ M, then B
′
returns the set of objects that have
the attributes of B in common. Thus, B
′′
returns the
set of attributes common to all objects that have the
attributes of B in common; consequently, B ⊆ B
′′
.
As an example, using Table 1, objects
A = {Tweet2, Tweet3}, when submitted
to the operator described above, will re-
sult in A
′
= {Bolsonaro, Elections}. So
{{Tweet2, Tweet3}, {Bolsonaro, Elections}} is
a concept. All concepts found from Table 1 are
displayed in Table 2.
Table 2: Existing concepts in the formal context of Table 1.
Objects Attributes
{Tweet 1, Tweet 2,
Tweet 3, Tweet 4}
{}
{Tweet 4} {Fake News}
{Tweet 1, Tweet 3} {Lula}
{Tweet 2, Tweet 3} {Bolsonaro, Elections}
{} {Lula, Bolsonaro,
Fake News, Elections}
In Table 2 there is a concept with an empty at-
tribute set and a concept with an empty object set.
They are called infimum and supremum, respectively.
2.3 Triadic Concept Analysis
Initially, Triadic Concept Analysis (TCA) was de-
fined by Lehmann and Wille (Lehmann and Wille,
1995) which extends FCA, but a new dimension was
added (Wille, 1995).
Formally, a triadic context is given by the quadru-
ple (K
1
,K
2
,K
3
,Y ), where K
1
, K
2
and K
3
is defined by
the sets and Y the relation of the K
1
, K
2
and K
3
, i.e.,
Y ⊆ K
1
x K
2
x K
3
, the elements of K
1
, K
2
, and K
3
are called (formal) objects, attributes, and conditions,
respectively, and (g, m, b) ∈ Y is read: the object g has
the attribute m under the condition b. An example of a
triadic context is represented in Table 3. This example
shows the dataset with 3 dimensions: Users, Months
WEBIST 2024 - 20th International Conference on Web Information Systems and Technologies
168

and Terms. We have the Users/ID {1,2,3,4} as ob-
jects, Months {July, August} as attributes and Terms
{Lula,Bolsonaro,Fake News,Elections} as conditions.
Implications are dependencies between elements
of a set obtained from a formal context. Given the
context (G, M, I) the rules of implication are of the
form B → C if and only if B, C ⊂ M and B
′
⊂ C
′
(Gan-
ter et al., 2005). An implication rule B → C is con-
sidered valid if and only if every object that has the
attributes of B will also have the attributes of C.
We can define rules, as follows: r : A → B(s, c),
where A, B ⊆ M and A ∩ B = ∅. We can also de-
fine the support of the rules, which is defined by
s = supp(r) =
|A
′
∩B
′
|
|G|
and the confidence of the rules,
which is defined by c = con f (r) =
|A
′
∩B
′
|
|A
′
|
(Agrawal
and Srikant, 1994).
Table 4 shows two existing rules in the context of
Table 1. The rule Bolsonaro → Elections has 50%
support because this rule happens in 2 tweets, out of
a total of 4 tweets. The confidence is 100%, since
whenever a tweet has Bolsonaro it also has Elections.
When a rule has 100% confidence, such as the rule
Bolsonaro → Elections, it is called an implication.
2.4 Database Processing
Textual databases need to be pre-processed before be-
ing analyzed. The steps performed in this work are
the following: N-Gram, stop word removal, and Reg-
ular Expression.
• N-Gram: is a contiguous sequence of n items from
a given sample of text. The items can be letters or
words that are in sequence on a text sample;
• Stop Word Removal: consists of removing words
such as articles and prepositions, as these words
are not significant for textual analysis;
• Regular Expression: a technique to determine a
pattern in a text sample. It is used to find a group
of words that need to be replaced or deleted.
The steps described above were applied through
the Python package Natural Language Toolkit
(NLTK). The NLTK package has a list of stop words,
such as “the”, “a”, “an”, and “in”, so those words in
the list are removed from the database being prepro-
cessed, as these words are not meaningful to the anal-
ysis. This allows the database after pre-processing
to have a reduced size and also reduces the analysis
time (Contreras et al., 2018).
An n-gram is a contiguous sequence of n items
from a given sample of text. The items can be letters
or words that are in sequence in a text sample. An
n-gram of size 1 is referred to as a unigram and does
not consider other words that are in sequence. Size
2 is a bigram, and size 3 is a trigram, meaning that a
group of three words are in sequence in a text sample
(Roark et al., 2007). Table 5 shows an example of
bigrams and trigrams found in a text sample.
3 LITERATURE REVIEW
Several works are relevant to the context of this study.
These include works related to topic detection in so-
cial networks, topic evolution, and the classification
of textual content. These works will be described in
the following paragraphs.
Zhang et al. (Zhang et al., 2012) detail how the de-
tection of topics on the Internet is a challenge because
the information produced on the Internet is succinct
and does not adequately describe the real context be-
ing addressed. To solve this characteristic of the infor-
mation produced on the Internet, the authors used the
technique of pseudo-relevance feedback, which con-
sists of adding information to the data being analyzed.
With this strategy, the authors were able to en-
hance the quality of information available on the In-
ternet, refining the context with which this informa-
tion is associated. Consequently, they identified the
trends within this information that are likely to be-
come more prevalent on the Internet in the future.
This research also aims to detect topics within the
content produced on Twitter. However, the pseudo-
relevance feedback technique was not applied, as the
authors specifically analyzed tweets related to one
context—the Brazilian presidential election.
Cataldi et al. (Cataldi et al., 2010) used the topic
detection technique to identify emerging topics in the
Twitter community. The authors were able to carry
out the identification considering that if the topic oc-
curs frequently in the present and was rare in the
past, and thus characterized them as emerging. To
enhance the strategy addressed, an analysis of the au-
thors of these emerging topics was carried out through
the Page Rank algorithm, to ensure that the emerging
topic is not present only in some bubble of the Twitter
community. Finally, a graph was created that con-
nects the emerging topic with other topics that are re-
lated to it, and that therefore have a greater chance of
becoming emerging topics as well. Unlike the work
described above, this research aims to use topic detec-
tion to analyze how these topics change over time.
Our results align with findings from previous stud-
ies, such as Cataldi et al. (Cataldi et al., 2010) on topic
detection in social media, but provide a unique con-
tribution by focusing on temporal evolution. In con-
trast to studies like Zhang et al. (Zhang et al., 2012),
Temporal Analysis of Brazilian Presidential Election on Twitter Based on Formal Concept Analysis
169

Table 3: Triadic formal context sample.
July August
ID Lula Bolsonaro Fake News Elections Lula Bolsonaro Fake News Elections
1 X X X
2 X X X
3 X X
4 X X X X
Table 4: Example of supported and trusted rules.
Rule Support Confidence
Bolsonaro → Elections 50% 100%
Lula →
Bolsonaro, Elections
25% 50%
which used pseudo-relevance feedback, our approach
focused on the context of the Brazilian presidential
election, providing more targeted insights.
Dragos¸ et al. (SM. et al., 2017) present an ap-
proach that investigates the behavior of users of a
learning platform using FCA. The log generated by
the platform contains information about the actions
that each student performs on the platform. Thus, the
log allows the identification of the profiles of students.
The use of FCA by Dragos¸ et al. occurs to con-
sider the instant of time that the actions are performed
by the students. It is relevant to profile students to
understand whether they are performing actions late,
early or on time. Therefore, FCA can be considered
as an alternative to study temporal events.
Cigarr
´
an et al. (Cigarr
´
an et al., 2016) used FCA
to group tweets according to the topics found. By us-
ing FCA, the work still manages to obtain a concep-
tual grid of the topics found, obtaining a hierarchi-
cal view of the topics, which is a differential to other
techniques. The proposal was among the best results
of the RepLab 2013 forum, proving the effectiveness
of FCA for the topic detection challenge.
Arca et al. (Arca et al., 2020) propose an approach
to suggest tags (meaningful human-friendly words)
for videos that consider hot-trend subjects, ensuring
the video receives more visibility by being related to
a trending subject. The original tags are inserted man-
ually, and these tags serve as input for the algorithm,
which matches them with hot trend subjects. Our pro-
posed method also identifies meaningful words, the
difference is that our input is tweets, and then ana-
lyzes how these words vary over time.
4 METHODOLOGY
This section presents our methodology to achieve the
proposed objectives. For this, the steps presented in
the sections below were performed.
4.1 Creation of the Dataset of Tweets
About the Presidential Election
To create the dataset, a Python script was used to run
daily and collect the most relevant tweets of the day,
using the filter provided by the Twitter API. This filter
ensures that only tweets with significant reach within
the social network are returned. The script collected
3,634 tweets during the period from July 23, 2022, to
September 8, 2022. The extended collection period
was a differentiator in achieving better results.
The dataset was created using the Twitter API
with filters applied to retrieve tweets in Portuguese,
specifically from Brazilian users. Only tweets con-
taining keywords Lula, Bolsonaro, and Elections
were considered. Additionally, the collection focused
on tweets with significant reach, such as those with a
minimum of 100 retweets or likes.
The challenges encountered during this process
include limitations imposed by the Twitter API, re-
stricting the number of requests that can be made
by each user. Currently, these limitations are even
greater, as it is not possible to use the Twitter API for
free, which hinders the reproducibility of this work.
After that the preprocessing of the database was
necessary, transforming the textual content of tweets
into a list of words that will be analyzed. To perform
the task, the techniques described in the Database Pro-
cessing section were used. Therefore, the final result
consists of a database that includes a tweet identifier,
the topic extracted from the tweet, and the timestamp
indicating when the tweet was published.
4.2 Reducing Formal Context
After the preprocessing stage, it is necessary to reduce
the formal context to eliminate topics that are not sig-
nificant because they are associated with only a few
tweets. Thus, in addition to removing attributes from
the formal context, this stage helps reduce the number
of concepts found, facilitating the final analysis of the
work, which is the validation of the concepts found.
To select topics that are present in the largest pos-
sible number of tweets, a Python script was utilized.
This script counted how many tweets each term was
present in and sorted these topics, listing first those
WEBIST 2024 - 20th International Conference on Web Information Systems and Technologies
170

Table 5: Example of bigram and trigram.
Text Sample Bigram Trigram
Topics change over time {Topics change}
{change over}
{over time}
{Topics change over}
{change over time}
with the highest number of appearances. The top-
ics with the highest number of appearances in tweets
were then chosen to compose the formal context.
4.3 Lattice Miner
With the topics already defined, it is necessary to pre-
pare the files that will be used as input for the Lat-
tice Miner tool. This tool which was developed at the
University of Qu
´
ebec (Missaoui and Kwuida, 2011)
will be used for constructing, visualizing, and manip-
ulating contexts. The tool reads files in JSON format
to generate the formal context. To accomplish this, a
Python script was created to generate these files in the
expected format for the tool.
Once the formal context is constructed, it becomes
possible to extract implication rules for analysis, thus
producing the expected results. The implication rules
are provided in XML files and were analyzed manu-
ally, as few rules were generated due to the low sup-
port they showed. In practical terms, a search was
conducted for rules that exemplify events or senti-
ments of users towards a politician.
5 RESULTS
To analyze the presidential election, the tweets were
collected between the period of 07/23/2022 and
09/08/2022. The tweets were obtained through the
Twitter API using the keywords Lula, Bolsonaro, and
Elections. The Twitter API filter was applied to re-
trieve only popular tweets, which reduced the total
number of tweets. However, these tweets generated
significant engagement on the social media platform.
After applying NLP techniques and selecting N-
Grams with meaning, the following attributes were
obtained for the formal concept: Alexandre de
Moraes, good versus evil, elections, Bolsonaro, Lula,
democracy, electoral research, Senator Rog
´
erio Car-
valho, secret budget, Bolsonaro no Flow, armed
forces, fake news, first round, Guilherme de P
´
adua,
genocidal, former convict, Jornal Nacional, corrup-
tion, vote, president, interview, cash, purchased prop-
erties, nursing salary floor, Fachin, and suspends arms
decree.
The condition of the formal context created is the
period in which the tweet was published. The period
used was 4 days, resulting in a total of 12 conditions.
As an example, Table 6 displays a sample of the gen-
erated formal context.
With the obtained formal context, it was possi-
ble to generate implication rules and analyze how the
generated rules reflect the events of the presidential
election in Brazil. Table 7 provides details on the gen-
erated rules.
The first implication rule to be discussed in this
study is composed of the terms “Elections” and “Fake
news”. This rule was observed during a crucial pe-
riod that coincided with the announcement of poten-
tial penalties for candidates spreading fake news dur-
ing the electoral period, between 08/08 and 08/15.
During this time, there was also a significant increase
in the scrutiny of the effectiveness of social media
in detecting and removing fake news, aiming to pre-
vent users from being misinformed. It is believed that
these events were reflected in Twitter discussions, jus-
tifying their identification in our study.
Furthermore, it is important to highlight that the
spread of fake news can have serious consequences
for democracy, such as interference in the electoral
process and manipulation of public opinion. For this
reason, the fight against fake news has become a
global concern, and the analysis of such data can con-
tribute to the understanding and combat of this grow-
ing phenomenon.
Analyzing the confidence metric among rules re-
sulting in fake news with the antecedents “Elections”
and “Bolsonaro”, it is possible to observe an interest-
ing trend. It is noted that the rule with the antecedent
“Bolsonaro” has higher confidence compared to the
rule with the antecedent “Elections”. This difference
in confidence suggests that the name of the presiden-
tial candidate, Jair Bolsonaro, was more strongly as-
sociated with fake news than the electoral context it-
self. This highlights that social media users had a dis-
tinctive perception that candidate Bolsonaro was in-
volved in the dissemination of misleading informa-
tion. The association between the name of a political
candidate and the spread of fake news may have influ-
enced public perception and sparked heated debates
and discussions on social media during the election
period.
In Table 7, the second identified implication rule
Temporal Analysis of Brazilian Presidential Election on Twitter Based on Formal Concept Analysis
171

Table 6: Formal context sample.
07-23-2022 - 07-26-2022 07-27-2022 - 07-30-2022
ID Lula Bolsonaro Genocidal Former convict Lula Bolsonaro Genocidal Former convict
1 X X
2 X X
3 X X
4 X X
Table 7: Implication rules that varied over time.
Time Period Antecedent Consequence Support Confidence
08-08-2022 to 08-11-2022 Elections Fake news 4% 13%
08-12-2022 to 08-15-2022 Elections Fake news 2% 16%
08-08-2022 to 08-11-2022 Lula First round 2% 16%
08-16-2022 to 08-19-2022 Lula First round 8% 26%
08-16-2022 to 08-19-2022 Bolsonaro Fake news 4% 33%
09-01-2022 to 09-04-2022 Bolsonaro Fake news 3% 22%
08-20-2022 to 08-23-2022 Lula Former convict 4% 25%
08-28-2022 to 08-31-2022 Lula Former convict 2% 14%
08-24-2022 to 08-27-2022 Lula Interview 2% 25%
08-28-2022 to 08-31-2022 Lula Interview 2% 7%
08-28-2022 to 08-31-2022 Bolsonaro
Cash,
Purchased properties
2% 10%
09-01-2022 to 09-04-2022 Bolsonaro
Cash,
Purchased properties
3% 22%
is composed of the terms “Lula” and “First round”.
Temporal analysis reveals that this rule occurred in
two distinct phases: the first between 08/08 and 08/11
and the second between 08/16 and 08/19. At the time,
the possibility of Lula winning the election in the first
round was a recurring topic in the media, as electoral
polls indicated that he was close to reaching 50% of
valid votes.
It is interesting to note how the support for this
rule increased significantly, going from 2% to 8%,
reflecting the growing support for Lula’s candidacy
and the public’s interest in this electoral scenario. It’s
worth highlighting that the increase in support is di-
rectly related to the evolution of electoral polls, which
showed the increasing preference of voters for the for-
mer president. On August 18th, a poll indicated that
Lula reached 51% of valid votes, further fueling the
discussion about the possibility of his victory in the
first round.
Therefore, it is possible to observe how the sup-
port metric can be useful for evaluating the evolu-
tion of discussions on the social network, and tracking
the rise of topics that become increasingly relevant to
users. This analysis is crucial for understanding the
impact of public opinion on elections and the con-
struction of candidates’ images.
The third identified implication rule in Table 7 is
composed of the terms “Bolsonaro” and “Fake news”.
It is interesting to note that this rule appeared in dis-
persed periods, first between 08/16 and 08/19 and
then between 09/01 and 09/04. The fluctuation in
the occurrence frequency of this rule can be explained
by two relevant events related to Bolsonaro and fake
news.
The first event occurred in mid-August when news
broke that a group of businessmen was allegedly or-
chestrating a coup in favor of President Bolsonaro.
This event generated significant media and social me-
dia attention, with Bolsonaro dismissing the news as
fake. This could have influenced the occurrence of
this rule in the first analyzed phase.
The second event that may have driven the oc-
currence of this rule took place in early September
when the Superior Electoral Court (TSE) fined Presi-
dent Bolsonaro for spreading fake news linking Lula
to the First Capital Command (PCC). This false news
circulated widely on social media, sparking debates
and discussions about the spread of misinformation
and its impact on elections. This event could have
contributed to the occurrence of the rule in the second
analyzed phase.
This implication rule illustrates how a particular
theme can emerge on the social network, be discussed
for a period, and then temporarily disappear, only to
resurface at another time. This highlights the volatil-
ity of social media and the dynamic nature of dis-
cussions that take place within them. Understanding
these dynamics is crucial for the analysis of public
WEBIST 2024 - 20th International Conference on Web Information Systems and Technologies
172

opinion and electoral campaign strategies.
The fourth identified implication rule in Table 7
is particularly interesting as it reveals the strong po-
litical polarization in Brazil. This rule is composed
of the terms “Lula” and “Former convict” and was
observed shortly after a televised debate among presi-
dential candidates in which Jair Bolsonaro referred to
candidate Lula as a “former convict”.
This act sparked a heated discussion on social me-
dia about whether Lula could be called innocent af-
ter events that discredited Operation Car Wash, the
largest corruption investigation in the country’s his-
tory. While Bolsonaro’s supporters applauded the
provocation, Lula’s supporters reacted with indigna-
tion, accusing the current president of attempting to
tarnish the image of the former president and leader
of the Workers’ Party.
The implication rule “Lula” and “Former convict”
illustrates how social media can be used as tools for
the propagation of political discourse and how polit-
ical disagreements can lead to polarized and heated
discussions on the internet.
It is interesting to note that the fourth identi-
fied implication rule in Table 7 showed a confidence
of 25% between 20 and 23 August, making it the
second-highest confidence found in the analysis. This
highlights the strong association between the name
of the candidate Lula and the term “former convict”,
suggesting that Jair Bolsonaro’s strategy was effective
in linking Lula to his past as a convicted individual in
the eyes of the law.
This result indicates how political rhetoric can in-
fluence discussions on social media and how polariza-
tion can lead to the widespread dissemination of bi-
ased information. Furthermore, this implication rule
emphasizes the importance of sentiment and opinion
analysis on social media to understand how political
rhetoric can influence public opinion and shape politi-
cal discourse around certain topics and public figures.
The fifth identified implication rule in Table 7 is
composed of the terms “Lula” and “Interview”. The
latter term refers to the interview conducted by Jor-
nal Nacional of TV Globo with all presidential can-
didates. It is interesting to note that this interview
took place on August 25, and yet the subject contin-
ued to be discussed on Twitter until August 31. This
fact demonstrates the relevance of the interview for
platform users, who engaged in discussions about the
candidates’ performance in that event.
In the specific case of the term “Lula”, it can be
inferred that the candidate’s participation in the inter-
view generated even greater interest among users who
discussed his performance and the ideas presented.
This highlights the importance of traditional media
in shaping public opinion and how social media plat-
forms can amplify and prolong the impact of these
events in society. Additionally, the persistence of the
discussion about the interview on Twitter also demon-
strates how social media can be used to monitor and
analyze public engagement regarding significant po-
litical events.
It is notable how the confidence metric of this spe-
cific rule significantly decreased in the second ana-
lyzed period. This drop in confidence indicates that,
after a few days from the mentioned interview, new
terms and topics related to candidate Lula started to
gain relevance in the discussion. This highlights the
volatility of the social network, where the discussed
themes and topics can change rapidly within a short
period.
This agile dynamic of the social network reflects
the ephemeral nature of conversations and the rapid
dissemination of information. In a matter of days,
other events, statements, or more recent occurrences
can capture the users’ attention and shift the focus to
more recent topics. This observation reinforces the
importance of continuously monitoring discussions
on social media to gain an updated understanding of
the landscape of opinions and topics of interest.
The sixth identified implication rule in Table 7 is
very relevant for understanding the Brazilian politi-
cal context. It is composed of the terms “Bolsonaro”,
“Cash”, and “Purchased properties” and is related to
news articles published by the Brazilian media that
highlighted many of the properties purchased by the
Bolsonaro family were acquired in cash.
This fact raised questions from Jair Bolsonaro’s
opponents, who sought to understand the reason be-
hind the cash purchases and the origin of this money.
The implication of these terms shows how issues re-
lated to ethics in politics are relevant to Twitter users,
who aim to discuss and comprehend the implications
of such actions by politicians.
It is noticeable how the analysis of topics dis-
cussed on Twitter can provide valuable information
for presidential campaigns. By understanding which
topics are generating more interest and discussion
among social media users, campaigns can adjust their
communication strategies and act more effectively to
win over the electorate. Additionally, the analysis
can also help campaigns identify weaknesses in their
strategies and improve them.
In summary, the analysis of topics on Twitter can
be an important tool for electoral campaigns to con-
nect with the electorate and better understand their
concerns and interests.
Temporal Analysis of Brazilian Presidential Election on Twitter Based on Formal Concept Analysis
173

6 CONCLUSIONS
This paper presents an approach to analyze topics
discussed on the Twitter social network. To iden-
tify these topics, the Twitter API is used to extract
tweets, and NLP is applied to find topics within the
tweets. Finally, FCA was utilized to generate impli-
cation rules between these topics, providing metrics
such as support and confidence.
With the generated implication rules and metrics,
it was possible to compare how Twitter users were
affected by events occurring at the time, such as the
Brazilian presidential election.
The Brazilian presidential election was evaluated
with a focus on the two most relevant candidates, Lula
and Bolsonaro. The analysis was conducted over a
month and a half, allowing topics that were being dis-
cussed to be forgotten and then remembered by users
of the social network, as was the case with the topics
“Bolsonaro” and “Fake news”.
The results demonstrate that it is possible to as-
sess how users are reacting to events related to the
election, information that is relevant for presidential
campaigns. Campaigns need metrics to evaluate the
impact of their actions.
As future work, there is a plan to automate the
entire process so that data can be extracted from the
social network and analyzed automatically, generat-
ing metrics on the topics discussed at the same mo-
ment. This is relevant for analyzing this informa-
tion at the same time it is being discussed on the
social network. Also, could extend this analysis by
incorporating sentiment analysis to better understand
the emotional tone of discussions. While our current
study focuses on topic evolution, adding a sentiment
dimension could reveal more nuanced insights into
public opinion. Future work could focus on automat-
ing the rule analysis process using machine learning
techniques, which would allow for a more extensive
and unbiased evaluation of the generated rules and
improve the overall efficiency of the method.
ACKNOWLEDGEMENTS
The authors thank the Pontif
´
ıcia Universidade
Cat
´
olica de Minas Gerais – PUC-Minas and
Coordenac¸
˜
ao de Aperfeic¸oamento de Pessoal de
N
´
ıvel Superior — CAPES (CAPES – Grant PROAP
88887.842889/2023-00 – PUC/MG, Grant PDPG
88887.708960/2022-00 – PUC/MG - Inform
´
atica,
and Finance Code 001). The present work was also
carried out with the support of Fundac¸
˜
ao de Amparo
`
a Pesquisa do Estado de Minas Gerais (FAPEMIG)
under grant number APQ-01929-22.
REFERENCES
Agrawal, R. and Srikant, R. (1994). Fast algorithms for
mining association rules in large databases. In Pro-
ceedings of the 20th International Conference on Very
Large Data Bases, VLDB ’94, pages 487–499, San
Francisco, CA, USA. Morgan Kaufmann Publishers
Inc.
Arca, A., Carta, S., Giuliani, A., Stanciu, M., and Refor-
giato Recupero, D. (2020). Automated tag enrichment
by semantically related trends. In Automated Tag En-
richment by Semantically Related Trends, pages 183–
193.
Cataldi, M., Di Caro, L., and Schifanella, C. (2010). Emerg-
ing topic detection on twitter based on temporal and
social terms evaluation. In Proceedings of the Tenth
International Workshop on Multimedia Data Mining,
MDMKDD ’10, New York, NY, USA. Association for
Computing Machinery.
Cigarr
´
an, J.,
´
Angel Castellanos, and Garc
´
ıa-Serrano, A.
(2016). A step forward for topic detection in twitter:
An fca-based approach. Expert Systems with Applica-
tions, 57:21–36.
Contreras, J. O., Hilles, S., and Abubakar, Z. B. (2018).
Automated essay scoring with ontology based on text
mining and nltk tools. In 2018 International Confer-
ence on Smart Computing and Electronic Enterprise
(ICSCEE), pages 1–6.
Ganter, B., Stumme, G., and Wille, R., editors (2005). For-
mal Concept Analysis: Foundations and Applications.
Springer.
Ganter, B. and Wille, R. (1999). Formal concept analy-
sis: mathematical foundations. Springer, Berlin; New
York.
Lehmann, F. and Wille, R. (1995). A triadic approach to
formal concept analysis. Conceptual structures: ap-
plications, implementation and theory, pages 32–43.
Missaoui, R. and Kwuida, L. (2011). Mining triadic asso-
ciation rules from ternary relations. Formal Concept
Analysis, pages 204–218.
Roark, B., Saraclar, M., and Collins, M. (2007). Discrimi-
native n-gram language modeling. Computer Speech
Language, 21(2):373–392.
SM., D., C., S., and S¸otropa DF. (2017). An investigation of
user behavior in educational platforms using temporal
concept analysis. In ICFCA 2017.
Wille, R. (1995). The basic theorem of triadic concept anal-
ysis. Order, pages 149–158.
Zhang, J., Liu, D., Ong, K.-L., Li, Z., and Li, M. (2012).
Detecting topic labels for tweets by matching features
from pseudo-relevance feedback. In Proceedings of
the Tenth Australasian Data Mining Conference - Vol-
ume 134, AusDM ’12, page 9–19, AUS. Australian
Computer Society, Inc.
WEBIST 2024 - 20th International Conference on Web Information Systems and Technologies
174