Abordagem computacional para avaliação automática da qualidade da argumentação na dimensão retórica de tweets no domínio da política brasileira

• Autor(a) principal: Silva, Cássio Faria da
• Data de Publicação: 2023
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Repositório Institucional da UFSCAR
• Texto Completo: https://repositorio.ufscar.br/handle/ufscar/18253
• topic: Avaliação da qualidade da argumentação; Tweet; BERT; Política brasileira; Processamento da língua portuguesa; CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::METODOLOGIA E TECNICAS DA COMPUTACAO

Research in the area of argumentation, inherent to human beings and essential for both spoken and written communication, dates back to the 4th century BC. The argument is multidisciplinary and covers several fields of research, including computer science. Communication has evolved to social networks, which are a considerable source of argumentative texts in various domains, such as politics. The automatic evaluation of arguments is the subject of recent studies in the area of Natural Language Processing and computational models that seek to perform this task are being proposed, especially based on algorithms based on Support Vector Machines and Deep Learning. In parallel, corpus of argumentative texts in English are being produced. As a way of contributing to research related to the assessment of the quality of argumentation in Portuguese, this work aims to propose, implement and validate a computational approach, for the automatic assessment of the quality of argumentation in the rhetorical dimension in tweets related to Brazilian politics. The approach involves developing a computational model, an annotated corpus with policy-related messages, and task-specific annotation guidelines. The studies carried out here showed that the most appropriate way to assess the quality of arguments in tweets related to Brazilian politics was using a neural model generated from the fine-tuning of BERTimbau. The proposed model was able to predict with 100% accuracy instances of the High quality argumentation class.