Fake News Detection Using Heterogeneous Information from Multimedia Content

In today’s digital world, most people prefer to read the news online, and they only find it accessible if the news includes images, text, audio, or video. In recent years, the news on social media platforms has been becoming increasingly crowded with multimedia reports and knowledge. The widespread availability of multimedia news on social media has contributed to the dissemination of fake information. In this paper we propose a Multimodal Cross Attention CNN Network (MCACN) method for multimodal fake news detection that learns the correlation between the textual and the visual content of the news. We employ the cross attention mechanism to fuse the features from the visual and the textual modalities. The attention mask for the visual features is completely dependent on the textual feature of the news and vice-versa. The findings of this research indicate that the suggested technique works better than the methods that just rely on lingual or visual information alone. Additionally, the proposed method provides a mechanism that can be used for detecting comparable disinformation operations on multiple contexts within the same social domain.