COVID-19 Fake News Detection Using Cross-Domain Classification Techniques

The recent pandemic has witnessed a parallel infodemic happening on social media platforms, leading to fear and anxiety within the population. Traditional machine learning (ML) frameworks for fake news detection are limited by the availability of data for training the model. By the time sufficient labeled datasets are available, the existing infodemic may itself come to an end. We propose a COVID-19 fake news detection framework using cross-domain classification techniques to achieve high levels of accuracy while reducing the waiting time for large training datasets to become available. We investigate the effectiveness of three approaches: Domain Adaptive Training, Transfer Learning, and Knowledge Distillation that reuse ML models from past infodemics to improve the accuracy in detecting COVID-19 fake news. Experiments with real-world datasets depict that Transfer Learning performs better than Domain Adaptive Training and Knowledge Distillation techniques.