Unveiling Truth Amidst the Pandemic: Multimodal Detection of COVID-19 Unreliable News

The prevalence of social media as a primary news source raises concerns due to the rapid spread of fake news. A significant majority of Twitter users (59%) and Facebook users (54%) rely on these platforms for their day-to-day news consumption, as observed by the PEW Research Center. This reliance extends to other social media platforms like Reddit, YouTube, and TikTok. The increasing dependence on social media for news has significant impacts, particularly in critical areas such as healthcare during the COVID-19 pandemic, election outcomes, emergency management, and public trust in institutions. To combat the detrimental effects of fake news, computational analysis techniques that incorporate multimodal features are crucial for effective detection and countermeasures. This study proposes a multimodal approach utilizing text embeddings from Fine-tuned BERT and image embeddings from CLIP to detect unreliable news. Experimental results on a ReCOVery COVID-19 dataset demonstrate the model's superiority over competitive baselines, particularly in detecting unreliable news. The findings highlight the potential of this approach in identifying and mitigating the spread of fake news. By combining text and image embeddings, this research offers a promising strategy for enhancing fake news detection capabilities and fostering trust in news dissemination on social media platforms.