A BERT-Based Semantic Enhanced Model for COVID-19 Fake News Detection

During the COVID-19 pandemic, COVID-19-related news keeps growing and spreading daily across social media platforms, including text, pictures, and videos. Meanwhile, fake news spreads widely on the Internet, preventing authoritative information from spreading and hindering the fight against the disease. To detect and recognize fake news, as well as to prevent its spread, effective detection models are urgently required. Text information is the most significant component of news content and is easy to be adopted by news consumers, so text-based fake news detection models are highly desirable. In this study, we propose a transformer-based semantic enhanced classification model for COVID-19 fake news detection. The model adds a semantic extraction module to the vanilla classifier to extract topic information from data samples as additional features to supplement text representations. Using k-fold cross-validation, we validate the model's performance on a publicly available COVID-19 fake news dataset, demonstrating its effectiveness and robustness. On evaluation metrics, the proposed model performs better than the vanilla model by more than 3%.