Research On Social Network Rumor Detection Technology Based On The Analysis Of Public Opinion Evolution Propagation

Online rumors have seriously affected the orderly and healthy development of cyberspace.It is important and urgent to find and curb the propagation of rumors in time.The rumor detection is to screen for suspected rumor events,but current research only roughly proposes the concept of early detection,and there is no detailed division of rumor detection stages yet;Moreover,in different periods of rumor detection,there are still problems such as low accuracy of early detection methods,the poor detection performance of models based on local propagation information,and a lack of strong interpretability and high-performance unsupervised detection methods.In response to the above issues,the paper conducts research on social network rumor detection technology from the perspective of public opinion evolution and dissemination.By introducing the public opinion dissemination cycle,rumor dissemination is divided into five stages,and research is carried out in three stages: the budding stage,the outbreak stage,and the elimination stage,corresponding to three types of scenarios: no dissemination information,local dissemination information,and global dissemination information.This article aims to identify rumors within a propagation cycle,with the aim of early detection,outbreak warning,and containment of resurgence.The main work includes:1.A rumor detection method for the budding stage based on binary information reconstruction is proposed.The current rumor detection in the budding stage relies too much on natural language processing technology,attaches importance to extracting information from the text to optimize the detection model,fails to deeply mine and effectively combine user information,and the detection accuracy is poor.The feature representation used for rumor classification needs further optimization.The "User-Tweet" binary information is optimized by performing crossinformation transmission in the constructed neighborhood information graph to solve the problem.Firstly,based on the neighborhood information system and two theoretically driven assumptions,neighborhood graph relationships for users and semantics are constructed.Then the user information and Semantic information are passed between two neighborhood information graphs,and the potential object relationships in the neighborhood graph are captured through the graph neural network.Finally,utilizing weighted integration,the binary information is reconstructed and used for rumor classification.Experiments on two real-world datasets,Weibo in Chinese and PHEME in English,show that the proposed method outperforms the comparative method in Precision,Accuracy,Recall,and F1.2.A rumor detection model for the outbreak stage based on the fusion of static spatiotemporal information is proposed.The existing static graph models only focus on the spatial structure or temporal propagation information and fail to combine them.The model’s detection performance is poor.The spatial structure information and temporal propagation information of rumor propagation are fused through the integration of neural network models to solve this problem.First,semantic information is extracted,and undirected and directed propagation graphs are constructed.Then,the spatial structure information of rumor propagation is extracted through undirected graph convolutional network,the temporal propagation information of rumor propagation is extracted through directed recurrent neural network,and different source node information jumps are used to enhance each other.Finally,static spatiotemporal information is fused through a weighted ensemble and used for rumor classification.Experiments on datasets such as Weibo and Twitter have shown that the proposed method outperforms multiple advanced comparative methods in Precision,Accuracy,Recall,and F1.At the same time,a new data collection indicator was adopted,which greatly reduced the training time of the model.Experiments on the newly collected data show that the proposed algorithm outperforms other comparative algorithms.3.A rumor detection method for the elimination stage based on neighborhood information transmission is proposed.During the elimination period,rumor detection methods pursue strong interpretability and high performance,and unsupervised detection methods have become the focus.However,the current unsupervised learning focuses on the deep learning model,which has poor interpretability and relies heavily on large-scale sample data,and a small number of traditional methods based on data correlation or prior knowledge also have the problem of low detection performance,lacking unsupervised methods with strong interpretability and high detection performance.Hence,the importance score of the object is optimized by performing information transmission in the constructed neighborhood information graph,and a propagation-based enhancement strategy is adopted to improve the performance.Firstly,the initial importance of the object is evaluated from the perspective of local and global heterogeneous distances,which serves as the initial information.Then,a neighborhood information graph is constructed based on the neighborhood relationships on the mixed data,information transmission between objects is performed,and the importance score is continuously updated by paying attention to the correlation between neighbors.Finally,a propagation-based enhancement strategy is adopted to optimize the scores and search for rumor targets through importance ranking.Experiments on two UCI datasets and Weibo real-world rumor dataset show that the proposed algorithm outperforms multiple comparison algorithms in terms of Top radio unsupervised metrics and that the enhancement strategy effectively improves the detection performance of unsupervised methods.