Banca de QUALIFICAÇÃO: ANDRÉ RICARDO RAUCCI

Metodologia de Identificação e Classificação de Conflitos em Redes Sociais Online

Methodology to identify and Classify Conflits at Online social Networks

With the advancement of technology in society as a whole, we notice that the Internet, online social networks, 

algorithms and data are increasingly present in people's lives [1]. 

With the emergence of the COVID-19 pandemic in early 2020, personal contact has become inconvenient and 

dangerous. Once social distancing became practically a premise, society looked for ways to accelerate digital 

transformation [102], including to solve communication problems that were previously solved in person [104], [105]. 

With all this digitalization, there were also many clashes on social networks (mainly Twitter), where each person 

or group carried out real battles to impose their narratives and opinions [16]. Twitter is one of the most popular 

social media platforms covering all types of content including health related texts. The Platform allows users to 

write short messages, called “tweets”, consisting of 280 characters and 140 characters before September 2017 [103]. 

Tweets are often adopted to share personal opinions, feelings, thoughts and activities. With over 500 million t

weets posted every day [110], Twitter has become a very valuable data resource for real-world insights. 

In the health domain, Twitter has also been adopted by users to share their personal health status, their experience 

with care and treatment options with other users with similar conditions/diseases and symptoms, as well as to share 

and seek health information more widely. interest, attracting the attention of clinical and biomedical researchers

 with the ultimate goal of improving patient outcomes .

Analyzing this phenomenon, the process of high polarization in social networks in political debates and vaccines 

(mainly linked to the subject of the Coronavirus), has resulted in a great social fragmentation, when two or more 

groups do not reach a consensus [3]. Therefore, an opportunity was identified to carry out a study of identification

 of polarizers in social networks by means of centrality techniques and to understand the reason for this polarization

 using artificial intelligence of clustering and sentiment analysis, in a timeline, in order to observe if people may or 

may not change their opinion along a timeline and why such people are able to influence a large group of followers. 

An important project called "Observatory of Internet Conflicts" was carried out by Universidade Federal do ABC,

 where we have some researchers and Master's and Doctoral students participating together. Each one making

 its contribution in relation to the analysis of the data. I had the opportunity and privilege to participate in this

 project, and my contribution was to carry out this work in a theoretical and practical way. With this, we started 

the construction of a project for the Identification of Conflicts in Social Networks (ICRS) that aims to offer a set 

of standards that help to identify conflicts and polarizing users in social networks. The idea is not only to identify 

conflicts and their users, but also to understand the dynamics and flow of how discussions are initiated and why 

some specific people become polarizing users, what are their profiles, and influence countless other polarized users 

with their narratives. Therefore, this work proposes a methodology that uses heterogeneous techniques in order to 

understand the dynamics that are established in a conflict. Experiments were carried out with several datasets 

(collected by the Vacinómetro do Observatório project) from the real world collected from the Twitter environment 

(in the year 2020), in order to enable the desired results of this research. All of this data is directly linked to 

discussions about the COVID-19 vaccine.

Therefore, an analysis of the data for the year 2020 (12 harvests or months) was carried out on the subject of the 

COVID-19 Vaccine. The analysis is separated by month, that is, twelve analyzes were performed. The challenge 

of the proposed methodology is the identification of polarizing users in the context of vaccine discussion through 

retweets, mentions and responses. Also, know if such polarizing users remain along the timeline through data 

science analysis and, then, perform a classification of these users through the texts of their tweets, in order to 

discover the reason for their polarization. Finally, also perform a sentiment analysis to support the conclusion 

of the results. Therefore, the purpose of this work is to propose a methodology (a set of data science techniques 

used together and in sequence), with the objective of presenting reliable results of conflict analysis of online social 

networks. With the application of all the techniques proposed in this work, we were able to identify the polarization

 of the main users, know the reason for this polarization (phenomenon) and also identify that some changed their 

minds along the timeline.