Multi-modal Chinese Fake News Detection

Abstract

The growth of fake news has been accelerated by the popularity of the Internet, creating a fertile ground for its dissemination. With the advent of diverse social platforms, fake news has evolved beyond textual content, extending to multimodal formats like images and videos. Consequently, there is an urgent need to develop fake news detection methods that are effective for the current multi-modal information landscape. This paper presents a model called MMCFND for detecting multimodal fake news in the Chinese context. The proposed MMCFND model leverages both textual and visual features extracted from news articles and accompanying images. To enhance cross-modal semantic understanding, image-text alignment learning and contrastive learning techniques are employed. Additionally, a hybrid expert system and cross-attention mechanism are incorporated to enhance detection performance in tasks involving multiple domains and modalities. The experimental results demonstrate the superiority of the proposed model compared to existing single-modal detection models. This showcases its effectiveness in tackling the challenges introduced by multi-modal fake news detection in the Chinese language. By combining textual and visual information, the model achieves improved accuracy and robustness in identifying fake news across various domains.