Análise de emoções em tweets de resposta a posts do ex-presidente do Brasil Jair Messias Bolsonaro

• Autor(a) principal: Assi, Fernanda Malheiros
• Data de Publicação: 2023
• 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/18488; https://github.com/LALIC-UFSCar/Analise-de-emocoes-em-tweets-na-politica-brasileira
• topic: Processamento de linguagem natural; Ciência política computacional; Detecção de emoção; Política brasileira; Twitter; CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::METODOLOGIA E TECNICAS DA COMPUTACAO

In recent decades, several computational advances have allowed more democratic access to the internet and, consequently, to social networks. The reach of digital politics is vast, and Twitter’s role in this scenario is undeniable. This platform became a significant political debate stage, where citizens, leaders, and institutions interact directly. In Brazil, this dynamic is especially relevant given the intense activity of former president Jair Messias Bolsonaro on the platform. However, manual analysis of such a large volume of data is impractical and error-prone. For such textual data to be transformed into information, it is necessary technologies to extract and process it. The NLP offers valuable tools for the automated analysis of large volumes of text. Among these tools, GoEmotions stands out, a machine learning model that can identify 27 categories of emotions in texts. Thus, this work used GoEmotions, adapted for Portuguese, to categorize the emotions in tweets in response to former president Jair Bolsonaro’s tweets to understand which emotions predominate in this context. The analysis of the results showed a substantial imbalance in the presence of emotions in the corpus, with “anger” being the most predominant emotion. In addition, a varied performance of the model in the identification of emotions was verified, with greater precision in the identification of “anger”, but presenting challenges with complex emotions such as “admiration” and “curiosity”, especially in contexts of irony or sarcasm. These results point to the potential of NLP models in analyzing emotions in political tweets while also highlighting the need for continuous improvements to deal with language nuances.