Detecção automatizada de desinformadores em grupos públicos do whatsApp utilizando redes neurais para grafos

Abstract

In recent years, the spread of false information on digital platforms has raised growing social concerns, particularly regarding its impact on public opinion, democratic processes, and public health. Among these platforms, WhatsApp stands out in the brazilian context as one of the main channels for the dissemination of disinformation, especially within large public groups. Although the literature on disinformation detection has made significant progress, studies specifically focused on identifying misinformers in this environment remain scarce. This work proposes an innovative approach for the automated detection of such agents, leveraging GNN applied to a real dataset collected from public WhatsApp groups. The methodology involves modeling user interactions as a graph, enabling the capture of structural and dynamic aspects of dissemination behavior. Through a comparative analysis of different GNN architectures: (GCN, GCNII, GraphSAGE, GAT e APPNP), the effectiveness of these approaches in identifying misinformers was evaluated. The results confirm the potential of GNN to capture both semantic and topological patterns, thereby strengthening disinformation mitigation strategies on WhatsApp and further establishing this approach in the field.