Detecção de fake news em redes sociais com o uso de redes neurais recorrentes, redes neurais gráficas e transformers

Abstract

Throughout history, the use of fake news with the intention of deceiving, promoting product sales, and influencing opinions has been there. With the advent of social media, the environment has become conducive to the dissemination of these fraudulent news stories, due to the intersection between stimuli and sharing, the popularization of content production without supervision, and the personalization achieved through the analysis of large volumes of data using advanced modeling techniques. In this context, traditional approaches to dealing with disinformation become insufficient. Based on recent research in the fields of neural networks and behavior, we seek to demonstrate that it is possible to identify and understand the circulation of fake news on social media using deep learning algorithms. This endeavor requires an understanding of different processes of signification, necessitating a theoretical framework that enables us to comprehend the nature and dynamics of these processes. A suitable foundation for developing this understanding is found in C. S. Peirce's theory of signs, in which processes of signification are conceptualized as semiosis, an irreducible and processual triadic relationship between signs, objects, and interpretants. Therefore, this study addresses the application of different neural network architectures, such as recurrent, graph, and transformer networks, for detecting fake news. Furthermore, it aims at understanding whether graph-based neural network models can more effectively capture the communication structure among social media users and be less susceptible to modifications of the writing patterns of fake news. Finally, the study examines whether a combination of linguistic and graph-based models may be more efficient than their isolated use in combating fake news