Detecção de fake news com processamento de linguagem natural e uso de modelos de linguagem (LLMS)

Abstract

This work presents a study on automatic fake news detection in Brazilian Portuguese using Natural Language Processing (NLP) techniques and language models. We used the Fake.Br corpus, which contains news labeled as true or fake, and adopted the BERTimbau model combined with additional context automatically generated by the Qwen2.5 language model. The methodology comprised data preprocessing, context generation, model fine-tuning, and binary classification. Results show that adding contextual information improved classification performance, reaching 98% accuracy, 0.98 precision, 0.98 recall, and 0.98 F1-score on the test set. These findings highlight the potential of integrating language models and advanced NLP techniques to combat misinformation in Brazilian Portuguese.