Intelligent Multimodal Deep Learning Framework for Fake News and Rumor Detection in Online Social Networks

The rapid growth of online social networks (OSNs) has greatly improved global communication and information sharing, but it has also accelerated the spread of fake news and rumors. This misinformation can threaten political stability, public health, and social harmony. Conventional fake news detection methods that rely on rule-based techniques or basic machine learning models are often unable to address the complex and evolving nature of online misinformation. To overcome these challenges, this study proposes an Intelligent Multimodal Fake News Detection System (IMFNDS) that integrates different types of information for more reliable detection. The proposed framework analyzes textual, visual, propagation, and metadata features to enhance accuracy. Transformer-based models such as BERT are used for text understanding, while Convolutional Neural Networks (CNNs) extract visual patterns from images and videos. In addition, Graph Neural Networks (GNNs) model how information spreads across social network communities. These diverse features are combined using an attention-based multimodal fusion mechanism, allowing the system to focus on the most relevant signals. An ensemble classification approach is further applied to increase robustness and reduce bias. Experimental evaluation shows that the proposed model performs better than unimodal methods, achieving higher accuracy, improved F1-score, and effective early detection of misinformation. Overall, IMFNDS offers a scalable and interpretable solution for identifying fake news and supporting trustworthy online information ecosystems.