Análise de polaridade e de tópicos em tweets no domínio da política no Brasil

• Autor(a) principal: Capellaro, Leonardo
• Data de Publicação: 2021
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: por
• Título da fonte: Repositório Institucional da UFSCAR
• Texto Completo: https://repositorio.ufscar.br/handle/ufscar/15138
• topic: PLN; BERT; Análise de sentimentos; Análise de tópicos; Português do Brasil; Sentiment analysis; Topic analysis; CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::METODOLOGIA E TECNICAS DA COMPUTACAO

The political scenario in Brazil is one of the busiest and most controversial in the last decade. With the advent of social networks, a new communication channel between voters and politicians was created, with users having a space to publish their opinions, beliefs and also follow the posts of other people. In this context, this work shows the performance of BERT models in the polarity (positive, negative or neutral) and topic analysis of tweets related to Brazilian politics. Experiments were carried out with tweets related to the 2018 elections achieving results (F1 = 96%) better than the ones obtained by previous works for the polarity classification. The qualitative analysis of the topics (the first one using BERTopic in this corpus) also resulted in promising insights.