Fake account detection with attention-based graph convolution networks

Abstract

Social networks are permeating all aspects of social life. As the number of users of social networks continues to expand and social relationships continue to enrich, the security risks faced by social network environments are becoming increasingly apparent. Most of the existing methods need to manually extract some descriptive features, which leads to the performance of the model depends largely on the quality of the extracted features. Aiming at the problem, this paper proposes an attention-based graph neural network, which uses the graph convolutional operator to capture the aggregation patterns in social networks automatically. We verified and discussed this proposed hypothesis on the cresci2017 dataset. Experimental results show that our attention based GNNs are better able to capture malicious account behavior than previous non-learning methods.