Rumor detection based on emotion aware and graph convolutional network in big data social networks

Traditional rumor detection methods tend to ignore the emotional features in posts and comments, and do not consider the extraction of key features from user comments and forwarding. This paper proposes a new rumor detection method for social networks, emotional perception rumor detection, combined with attention mechanism and graph convolutional network (EaA-GCN). First, a text sequence is constructed from the posts and comments of each event, and a multi-head attention mechanism is used to aggregate the features of emotion and text, so as to obtain the semantic features containing emotion tendency. Second, the propagation topology is constructed according to the propagation and forwarding relationship of the post. Combining the root node information enhancement and DropEdge method, the influence of root node is enhanced and false and irrelevant interactive information is suppressed, so as to obtain more effective communication structure features. Finally, EaA-GCN is proposed by multi-scale fusion of high-dimensional features extracted from different modules. A large number of experimental comparisons and performance analysis were carried out with nine models including DTC on three datasets of Weibo, Twitter15 and Twitter16. On Weibo dataset, EaA-GCN method was superior to baseline machine learning and deep learning detection methods such as DTC and PPC in terms of Acc. The Acc of the best baseline model VEVE-GCN was increased by 4.1% on Twitter 15 and Twitter 16 data sets. Compared with DTC, PPC and other methods, the EaA-GCN method had different degrees of improvement in Pre index. The experimental results verify the rationality and effectiveness of EaA-GCN method, and can be used in the field of general false information detection after further improvement.