Research On Social Network Rumor Detection Model Based On Multi-Feature Fusion

The rapid development of social networking platforms has gradually become a carrier for the breeding and widespread dissemination of rumors while providing information convenience for the public.This thesis uses the differences between rumors and real information in terms of text semantic features and communication structure features to build a rumor detection model.The specific research contents are as follows:(1)Most rumor detection methods use recurrent neural networks to extract the text semantic features of rumors.When the text sequence is too long,the ability of the model to extract rumor features will be reduced.In addition,most rumor detection models use a single graph convolutional network to extract propagation structure features,resulting in weak feature extraction capabilities.To address these issues,this thesis proposes an edge-learningbased multi-feature fusion rumor detection model(AEGCN).Firstly,the attention mechanism is used to mine the deep textual semantic features of rumors.Afterwards,a propagation structure graph is constructed.To enhance the model’s ability to extract structural features of rumor propagation,this thesis introduces an edge learning module based on the graph convolutional network to improve the performance of the model,and finally uses fusion features for detection.Experimental results show that the proposed AEGCN has high detection ability.(2)Most rumor detection methods extract the characteristics of rumors from two aspects of text semantics and propagation structure to achieve automatic classification of rumors,but most methods do not realize that false and irrelevant interactions in the propagation structure will reduce the accuracy to a certain extent.Moreover,previous rumor detection methods fail to effectively extract key clues from user comments in social networks.Aiming at these phenomena,this thesis proposed a rumor detection method(DA-GCN)that integrates dual attention mechanism and graph convolutional network.Firstly,the propagation graph is constructed,and the graph convolutional network is used to extract the propagation structure information of rumors and combined with the attention mechanism to suppress false and irrelevant interactions,so as to extract the anti-interference propagation structure features from the propagation graph.Second,to fully mine the clues in user comments,this thesis uses the attention mechanism to fuse the source Weibo and commentforward information and extract interactive text semantic features from them.Finally,the fusion features are used for detection.Experimental results show that the proposed DA-GCN achieves 94.4%,90.5% and 90.2% accuracy on the Weibo dataset and Twitter15 and Twitter16 datasets respectively,which proves that the proposed method is reasonable and effective.In summary,this thesis proposed a multi-feature fusion rumor detection model based on edge learning(AEGCN)and a rumor detection method that combines dual attention mechanism and graph convolution(DA-GCN),and makes corresponding comparisons for the two models.Theoretical analysis and experimental verification,the experimental results show that the detection model proposed in this thesis is reasonable and effective.