v3MFND: A Deep Multi-domain Multimodal Fake News Detection Model for Vietnamese.

Fake news become a critical problem on the Internet, especially social media. During the worldwide COVID-19 epidemic, social networking sites (SNSs) are primary sources to spread false news, which are incredibly difficult to detect and regulate them since they rapidly grow everyday. With multimedia technology advances, the content of social media news now is manifested via various modalities, such as text, photos, and videos. Approaches that learn the multimodal representation for detecting fake news have evolved in recent years. Additionally, there exist diverse content domains in news platforms. Exploiting data from these domains potentially solve the data sparsity problem as well as simultaneously boosting overall performance. In this paper, we propose an effective Deep Multi-domain Multimodal Fake News Detection model for Vietnamese, v3MFND for short. Extensive experiments on a real-life dataset reveal that v3MFND improves the performance of multi-domain multimodal fake news detection for Vietnamese considerably. An ablation study is also carried out to evaluate the role of each individual modality in the multimodal model.