A novel rumor detection method focusing on social psychology with graph attention network

The proliferation of online social media has led to an increase in the spread of rumors. Current rumor detection methods have not adequately considered the impact of social psychology and have neglected to integrate text features with other characteristics. This paper introduces a multi-feature fusion rumor detection model, Social Psychology-based Graph ATtention network (SPGAT), designed to enhance the accuracy of rumor detection. In this model, social psychological features are extracted using large language models, encompassing user emotion, social identity, group emotional resonance, and social influence. These features aim to deeply capture the essential attributes of rumors. Concurrently, a multi-head dynamic graph attention convolutional network is constructed. This network amalgamates complex structural features with essential features, thereby effectively capturing spatial propagation features and significant features while paying attention to the high-dimensional hidden features of rumors. Furthermore, a neural network is designed to comprehensively integrate the high- dimensional features of rumors, and rumor detection is achieved through a fully connected layer. Extensive experiments are conducted on three public datasets. Compared to the latest typical detection methods, the proposed method demonstrates certain advantages. Specifically, the accuracy and F1 score of rumor detection are improved by 5.87% and 4.4% on average on Weibo, PHEME 5, and PHEME 9 datasets compared to the latest baselines, respectively. Meanwhile, we verify and analyze the key role of socio-psychological characteristics in rumor propagation, which provides strong support for an in-depth understanding of the rumor propagation mechanism.