Spammer Detection Using Graph-level Classification Model Based On Graph Neural Network

The popularity of social networks makes spammers everywhere on the platform.First,Spams occupy the limited hardware resources and information transfer channels on the network.Their presence affects the efficiency of effective information transfer and seriously lowers the user experience.Second,spams pose multiple security risks to legitimate users.Social network platforms use technology to filter spammer.However,it is difficult to detect spammers effectively because of the various forms of interaction and the multi-dimensional nature of user relationships.In this paper,the graph-level classification model based on GNN is proposed to detect spammer on social network platform.The main work is as follows:(1)User features were extracted from three aspects: account information,social relation and social behavior.The sequence feature analysis is carried out on the records generated by the user’s operation behavior,and the graph feature analysis is carried out on the user’s social relation graph.Combined with the user account information,the feature selection is carried out based on PCA,and finally the node features of multi-feature fusion are obtained.(2)The principle of GNN and the graph-level classification model based on GNN are introduced.After integrating the social data,the social pattern graph of user is constructed.The nodes of the graph represent users and the edges represent social interaction information.The graph-level classification model based on GNN is built to detect spammers by using the dual concern mechanism of GNN on node features and graph-structure features.Bayesian optimization framework is constructed to adjust the hyperparameters of the model.(3)The graph-level classification model based on GNN is trained on Tagged dataset and the optimal model is established.The accuracy of 96.2% was obtained in the 10-fold cross validation experiment.In the comparison experiments,the performance of graph-level classification model based on GNN is superior to the boosted-tree-classifier and random-forest-classifier on composite indicator.The results show that graph-level classification model based on GNN can effectively detect spammers on social network platforms,and the effect is better than other current methods.