Detecção automática de fake news em tweets sobre as eleições brasileiras de 2022.

Abstract

Fake News are deliberately false or misleading information created and disseminated with the aim of deceiving the public. These news articles are often designed to resemble legitimate news. Their objective is to manipulate public opinion, spread misinformation, influence elections, generate controversy, or to have financial gains. With the advent of social media, people have started consuming news online as it is extremely simple, fast, and easily accessible. However, this has also led to an increase in the dissemination of fake news. In recent years, we have seen that elections and public opinion on important social issues have been influenced by the spread of fake news. They are created and spread rapidly, highlighting the urgent need for rapid detection. In this work, we propose a methodology for detecting fake news using deep neural networks, with a dataset of over 2 million tweets from the Brazilian presidential elections of 2022, labeled automatically by a weak supervision model, with F1-score of 98% on non-fake news tweets, and F1-score of 47% on tweets containing fake news.