FACT OR FAKE? COMPARING NEWS CLASSIFICATIONS USING DIFFERENT ARTIFICIAL NEURAL NETWORK ARCHITECTURES

Resumo

This article addresses the truthfulness news through artificial intelligence, applying different architectures of Artificial Neural Networks to classify news as false or true. It seeks to provide insight into machine learning in fake news detection, contributing to the understanding of the role of fact-checking. The article discusses the popularization of technology in the dissemination of digital information and emphasizes the importance of digital news in the construction of collective knowledge. It highlights concerns about the spread of fake news and showcases fact-checking initiatives on various platforms. The theoretical foundation explores concepts of Machine Learning, Natural Language Processing, and Artificial Neural Networks. The methodology details the use of the "Fake.Br Corpus" database and describes data preprocessing, its division, and the construction of classification models. The results indicate training, testing, and validation accuracies, as well as the comparison of ROC curves between models. The conclusion emphasizes the feasibility of news classification by artificial intelligence, with accuracies exceeding 90%. The study suggests continuing tests, focusing on the exploration of deep learning architectures with appropriate vectorization in the preprocessing stage.