Research And Implementation Of False Information Classification Based On Graph Neural Networks

With the rapid development of social media,people are increasingly inclined to obtain information from social media.However,the convenience of social media will also have a negative impact.There is almost no threshold for publishing information on social media,which makes social media the main source of dissemination of false information.The false information would have a bad influence on individuals,society,and even the country.Information such as users and comments on social media constitute a social network,which contains useful clues for detecting false information.However,considering the heterogeneity and complexity of social networks,some existing false information detection methods lack the use of the social network.Although some methods try to extract useful information from social networks to improve the effect of false information detection,the noise in the social network may mislead the detection results.In response to the above challenges,this thesis proposes two false information detection methods based on graph neural networks.The first method focuses on the task of rumor classification in false information detection.It starts from the disguise behavior of social network users.This method defines four common disguise behaviors and simulates them through the attention mechanism.Next,the graph neural network is used to extract the structural features of the social network.Then as many disguise schemes as possible are simulated through adversarial training so that the false information detector can resist the interference of the disguise behavior in the social network.The second method mainly focuses on the veracity classification task and the stance classification task in false information detection.It combines these two tasks through multi-task learning framework and makes the best use of the information in the social network.It firstly proposes a single-task framework based on social network and information content.Then it uses a shared layer with interactions to fuse the two single-task frameworks into a multi-task framework.Experimental results show that the above two methods have achieved significant improvements.Finally,this thesis integrates the above two methods to develop and implement a false information classification system based on graph neural networks.The system has a user-friendly visual operation interface.Users can perform data processing,model training and model test through interactive controls on the system interface.Users can adjust the relevant parameters according to the output of the system to obtain a better false information detection model.