Research On Rumor Detection Algorithm Based On Temporal And Interactive Features

In recent years,with the development of 5G and artificial intelligence technology,the mobile Internet has been further developed,which greatly facilitates people to obtain and exchange information on social media.However,due to the anonymity of the Internet,anyone can publish information on the Internet,which has led to the widely spread of unconfirmed rumors and disrupted social order.Many researchers have carried out related research on rumors on social media and proposed a variety of models to improve the performance of rumor detection.Social media rumors keep developing and changing with time and context,the features are not fully exploited and utilized,and the actual detection results are not ideal.Based on this situation,this study takes Weibo rumors as the research object and combines technical means such as contrastive learning,graph neural network,and data mining to study how to fully integrate the temporal and interactive features of rumors to further improve the effect of rumor detection.The main work and innovations of this paper are as follows:(1)This paper applies contrastive learning to rumor’s text representation,and a rumor detection algorithm model based on contrastive learning is proposed.For rumor data,text content is the main source of information,and the quality of text representation can affect the final detection results of the model.Based on the method of contrastive learning,the SimCSE model is used as a text extraction module.Different Dropout masks are used for text enhancement,positive and negative samples are constructed,and a better text representation model is obtained through unsupervised learning.The text representation obtained by the model is sent to the interactive feature extraction module based on the graph attention network for final rumor detection.(2)This paper considers both the temporal and interactive features of rumors,and a rumor detection algorithm model based on temporal and interactive features is proposed.Based on the Weibo rumor dataset,the time series data can be obtained by sorting the posts in the Weibo events according to timestamp.At the same time,according to the interaction relationship between the posts in the event,an interactive graph can be constructed to obtain the data with a graph structure.The number of posts of the event is counted,and the sequence data reflects the development and trend of the event,and then the discrete fourier transform is used to perform the time-frequency transformation on it to obtain the temporal features of the event;the parallel stacked graph attention network module is used to analyze the information of the post.The interactive features are modeled,and the interactive features are mined from multiple perspectives.Finally,the two features are combined for rumor detection,and good results are achieved.(3)This paper collects and constructs a new Weibo rumor dataset.The rumor detection research on Weibo has been carried out for a while,most of the research is oriented to the public rumor dataset,and this research also conducts related experiments based on this dataset.Considering that rumors will show different characteristics with the development of time,we designed and wrote data acquisition programs to collect and clean the data,and constructed a new Weibo rumor dataset.