Um estudo sobre a repercussão da eleição presidencial brasileira de 2022 no Twitter usando BERTopic

Abstract

One of the most reverberated themes throughout Brazilian society in 2022 was the presidential election. This election is considered by many to be one of the most important in history and moved discussions in virtually all media, whether online or not. In this work, we sought to identify the most discussed subjects under the theme of presidential elections on Twitter. To achieve this objective, Machine Learning and Natural Language Processing techniques were applied to a set of 1.4 million posts in Portuguese on the subject of the elections and specifically to the four main candidates for the Palácio do Planalto, in the period of 09/28/2022 to 10/05/2022. Techniques such as sentence embedding, clustering through HDBSCAN and TF-IDF, were used together with support from BERTopic, in order to group and automatically extract the most discussed topics. At the end of the process, the 12 most relevant topics were manually analyzed and compared with the news presented by the traditional media on the same days. Thus, it could be observed that BERTopic presented useful results for understanding Twitter discussions.