Discurso de ódio em redes sociais: uma abordagem automatizada para identificação em conteúdo em português

Abstract

Social media platforms play a pivotal role as means of human interaction in the contemporary era, providing platforms for communication and expression. However, the increasing prevalence of hate speech on these platforms poses significant risks to individuals and communities. Detecting and addressing hate speech is particularly challenging in languages like Portuguese due not only to its rich vocabulary, complex grammar, and regional variations but also to the scarcity of annotated datasets for this purpose. Based on this, this thesis presents the largest annotated dataset in Portuguese for automatic hate speech detection. The creation of this corpus is accompanied by an extensive literature review containing different definitions of hate speech, data availability, analyses of the peculiarities of hostile content propagation online in Brazil, and language model architectures. To assess the viability of the dataset, models with architectures representing the state of the art in the literature were used, with tests evaluating the use of BERT and GPT-2.