Análise temporal de tópicos em tweets de elites políticas durante a pandemia de COVID-19: avaliando diversidade, coesão, dissonância e coerência no discurso sobre vacinas

Abstract

With the increase in the use of social networks, important demands arise, such as analyzing their content to understand the influence these networks have on society’s behavior as a whole. Especially during critical moments like the Covid-19 pandemic period, it is essential to comprehend the factors that may have influenced the population’s behavior, contributing to or hindering containment efforts. In this study, we analyze the relationship between topics discussed in political elites’ tweets from Brazil within the context of vaccination, using natural language processing techniques, temporal topic modeling, and discourse analysis. The evaluated political elites were candidates for mayor throughout Brazil in 2020, providing insights into local political stances. We employed the Latent Dirichlet Allocation (LDA) model for topic determination. Based on the topic distribution generated by the LDA model, we calculated four discourse evolution metrics: diversity, cohesion, dissonance, and coherence. These metrics served as the foundation for the presented results and the hypotheses raised. The applied methodology, along with the obtained results, reveals that discourse evolution metrics contribute to understanding and provide a clearer view of how discourse evolves and its relationship with reality. Through this work, we were able to compare different entities, parties, or political groups in terms of the four calculated metrics, as well as identify the most relevant topics for each of them.