Research On Key Technologies For Judging The Authenticity Of Rumors On Social Networks

With the rapid development of the Internet,social network media is at the forefront of the development trend due to its unique nature,which also leads to the rapid development of online rumors.Due to the characteristics of fast information dissemination,low cost and low security on the Internet,it will increase the harm caused by false rumors to the society.Therefore,determining the authenticity of rumors from a large amount of information on social media has also become a hot research topic for researchers.Starting from the above research background,this paper will also study the key technologies of rumor authenticity based on machine learning and graph kernel algorithm.Usually,the authenticity detection of rumors mainly uses the rich language expression in the text content,the relevant user characteristics,and the characteristics that spread over time,but the text information can be maliciously tampered with,and a large number of users include fake users,which shows that this method itself has certain vulnerability.Therefore,this paper focuses on the mode of information dissemination,uses the graph kernel to extract the complex topological structure from the dissemination process of rumors,and then enhances the structure between dissemination nodes based on the user’s influence degree,and finally completely excludes any text information,User content and time information,and train a true and false rumor model based on propagation structure enhancement.Our research results prove that not only the authenticity of rumors can be detected through the propagation structure,but also the enhanced propagation structure model has high accuracy for the authenticity detection of rumors,and can identify the authenticity of rumors as early as possible to reduce the impact.Machine learning and deep learning technologies are currently effective methods commonly used in the field of rumors and their authenticity detection.However,the commonly used models also use CNN,LSTM and other models to process serialized data according to text content and user characteristics.Unserialized data such as structure diagrams cannot be processed.Therefore,in response to this problem,this paper proposes a rumor authenticity detection model based on D-GCN(Double Graph Convolutional Network),which realizes the differential processing of position detection and user influence,and extracts the structure graph and user graph for each conversation.The structure graph is constructed from source tweets and reply tweets,and the structural information of the graph and the position of each node can be obtained.The user graph is based on the users participating in the dialogue and the interaction information of the users in the dialogue,the user influence degree of each node can be obtained,and in the actual spread of rumors,the position of a node may be changed by the influence of other nodes.Then the outputs of the above two modules are combined and fully connected to obtain the classification results;the experimental results show that the rumor authenticity judgment model proposed in this paper has a high accuracy.