Fake News Detection Based on the Correlation Extension of Multimodal Information

Yanqiang Li, Ke Ji, Kun Ma, Zhenxiang Chen, Jin Zhou, Jun Wu

Abstract: Online social media is characterized by a large number of users that creates conditions for large-scale news generation. News in multimodal form (images and text) often has a serious negative impact. Existing multimodal fake news detection methods mainly explore the relationship between images and texts by extracting image features and text features. However, these methods typically ignore textual content in images and fail to explore the relationship between news and image texts further. We propose a new fake news detection method based on correlation extension multimodal (CEMM) information to solve this problem. The correlation between multimodal information is extended and the relationship between the extended image information and the news text is explored further by extracting text and statistical features from the image. This CEMM-based detection method consists of five parts, which can discover the relevant parts of news and optical character recognition (OCR) text and the features of fake news images and relevant parts of news text, and combine the information of the news itself to detect fake news. Experimental results proved the effectiveness of our approach.