Dual-Branch Graph Attention Model for Rumor Understanding and Propagation

With the widespread use of social media, although it has brought many conveniences, it has also intensified the spread of rumors. In recent years, graph neural networks have made significant progress in identifying the structural features of rumor propagation. However, more and more rumors adopt disguise strategies to evade the detection of the model. Inspired by the graph attention mechanism, this paper proposes a dual-branch graph attention model for rumor understanding and propagation, which integrates the characteristics of text content and propagation structure. The text content part constructs a graph network representation through the idea of self-supervised learning to capture the potential information in the text; while the propagation structure part adopts graph attention network modeling to fully explore the relationship between nodes and reveal the law of rumor propagation. In order to enhance the robustness of the model, we introduce the FGSM adversarial attack module in the training process to simulate the attack scenarios in practical applications, thereby improving the model’s ability to deal with various disguise strategies. Experimental results show that the proposed dual-branch graph attention model significantly outperforms the existing state-of-the-art rumor detection methods in experiments on two public datasets, showing stronger robustness and higher accuracy.