An Attributed Network Anomaly Detection Method Based on Dual Autoencoder Joint Embedding

Abstract

Anomaly detection in attributed networks has attracted a lot of attention in recent years. It is an important means to detect and find security anomalies in time and to take measures in advance to combat threats. Most existing methods ignore the complex interaction between network structure and node attributes, so it remains a challenging problem to better model the network structure information and the rich node attribute information to achieve the interaction between the two models. In this paper, we propose an attributed network anomaly detection method based on dual autoencoder joint embedding (DAJE). This method learns the embedding of both structural and attribute patterns separately by using structural and attribute autoencoders, and then reconstructs them while considering the consistency and complementarity of the structural and attribute information embedding. It takes into account the network structure and attribute interactions better than other methods, and its effectiveness is verified on three real-world datasets.