Research On Social Media Network Public Opinion Rumor Detection Metho

The vigorous development of social networks and the diversification of communication modes make information exchange more convenient and the amount of information data more abundant.People are no longer just recipients of information,but also creators of content.However,uneven user data brings great difficulties to how social supervision platforms can selectively and automatically mine user opinions,and the raging of online rumors will have a bad impact on the social trust system.The automatic rumor detection technology of network public opinion can quickly identify the true and false network public opinion through the effective learning of rumor characteristics,which plays a guiding role in predicting the trend of public opinion,and has important research significance.Focusing on the characteristics of network public opinion on social media platforms,this paper analyzes the role of various characteristics of public opinion in rumor detection.Based on new technologies such as graph convolutional network and graph attention network,this paper studies network public opinion rumor detection and analysis methods.Mainly complete the following research work:(1)Multi-task joint rumor detection method combined with commentsRumors in view of the present detection method based on the main news text itself,at the same time with the information such as user reviews,transmission characteristics to determine detection task in Chinese this feature,the user comments on the overall quality is not the problem of high considering comment on the effectiveness of the rumor is a kind of detection algorithm,the rumor to judge at the same time,considering the validity of user comments,Rumor detection is realized based on multi-task joint learning.Firstly,rumor detection task is taken as the main task,and user comment correlation detection task is taken as the auxiliary task.Then,gating mechanism and attention mechanism are adopted to filter and select valid user comment features.Finally,the experiment based on self-constructed microblog rumor data set shows that the filtering of user comments can not only improve the performance of rumor detection,but also realize the judgment of the quality of user comments.(2)Heterogeneous Graph Convolutional Network for Rumor Detection with MultiLevel Interactive Fusion and Graph ReconstructionAiming at the problems that current detection methods focus too much on contextual semantic information,seldom consider the impact of user communication on social media,and the problems of large differences in features in multi-feature fusion,relatively independent feature learning and feature error transfer,a new method based on semantic map and graph reconstruction rumor detection method with user propagation graph constraints.We first use an encoder-decoder framework to explore text semantic information and user propagation patterns with a multi-graph convolutional encoding module and a multi-graph reconstruction decoding module.Then a decision-level detection module is built to balance the global and local feature fusion process through multi-task learning.Experimental results on two common data sets demonstrate the superiority and advancement of the proposed method.