Rumor Detection Mechanism for Multi-Modal Information in Social Media

This paper proposes a novel approach for rumor detection in social media by integrating multi-modal information fusion. With the prevalence of false information and rumors on social platforms, an effective detection mechanism is essential. Our method leverages deep neural network architectures, including image captioning and recurrent neural networks (RNN) with attention mechanisms, to integrate text, image, and social features. We adopt two feature fusion strategies, early fusion and late fusion, to comprehensively integrate multimodal data. In addition, we design a multi-layer bidirectional recurrent neural network (BRNN) architecture to capture textual relations and improve classification accuracy. Experimental results on real datasets demonstrate the superiority of the proposed method, with an F-value of 0.89. This study promotes the advancement of social media rumor detection technology through effective multimodal information fusion and deep learning architectures. Future work will focus on optimizing model components and exploring advanced sentiment analysis methods for further performance improvements.