FakeWhatsApp.BR: detecção de desinformação e desinformadores em grupos públicos do WhatsApp em PT-BR

Abstract

In recent years, the large-scale propagation through social midia of false, misleading, or distorted information, i.e. disinformation, has become a serious social problem, harming organizations and individuals and negatively impacting democratic processes, economy, health and public safety. Thus, the study and development of methods for automatic detection of misinformation, as well as the detection of malicious users that spread misinformation, gained the attention of academia and industry. In Brazil and in several other countries, the mobile messaging application WhatsApp is one of the midia in which misinformation circulates the most. However, there are still few works in the literature that address the detection of misinformation in this specific scenario. In this dissertation, we propose the construction of FakeWhatsApp.Br: a dataset of messages obtained from public WhatsApp groups, containing propagation information (social and temporal), where messages shared more than once were labeled as containing or not misinformation. From this resource, we carry out a series of classification experiments using different machine learning techniques to detect messages with misinformation and misinformation spreaders. Classification methods based on natural language processing and user attributes were compared and discussed, analyzing the advantages and limitations of each approach and identifying the particularities and challenges of these problems. The results obtained in this work provide initial contributions to the study of these problems and point to future research in the context of misinformation on WhatsApp.