Spammer Detection Using Graph-level Classification Model of Graph Neural Network

摘要

The popularity of social networks has made spammers a ubiquitous presence on all platforms. Spammers occupy the limited hardware resources and information transfer channel, even pose multiple security risks to legitimate users. In this paper, the graph-level classification model of graph neural network is used to detect spammer on social network platform. We convert the behavior pattern of each user into a graph and extract the graph features for model training. Bayesian optimization framework is used in hyperparameters tuning. We obtain node features from account information, relation-graph and behaviour-sequence, and principal component analysis is used for feature selection. We have performed 10-fold cross-validation experiments on the Tagged dataset and get good results. The results show that graph neural network model has higher recognition accuracy for spammers than traditional classifiers, such as gradient tree classifier and random forest classifier.