Leveraging Dual GNNs and BERT for Context-Aware Rumor Detection in Social Networks

In the current digital era, the rapid expansion of social media has accelerated the spread of rumors, allowing misinformation to circulate rapidly and affect a wide audience. Detecting rumors is a challenging task due to their varying degrees of credibility and diverse categories. Current detection techniques generally focus on single features, such as textual information or sentiment analysis, while more advanced methods combine different data modalities, including text, images, and metadata, to improve accuracy. Some methods employ hierarchical structures; however, they often fail to comprehend the semantic and contextual nuances of the content and overlook the complete retweet propagation structure, impairing their effectiveness in detecting rumors. This paper proposes a Dual GNN BERT-based Model (DGBM) for rumor detection. Our model utilizes a Graph Convolutional Network (GCN) and a Graph Neural Network (GNN) to effectively capture contextual and semantic relationships between the words in source tweets and their retweets. We applied Gated-GNN to combine information from different levels and better understand how rumors spread. We conducted experiments on the Sina Weibo benchmark dataset, achieving an accuracy of 97% , which is 2% higher than the previous methods. This improvement demonstrates that our approach effectively captures local and global features of the propagation graph structure.