Relationship Prediction based Anomaly Detection in Heterogeneous Information Networks

Abstract

In heterogeneous information networks, there are many different types of nodes and different types of edges. Some homogeneous anomaly detection methods can't be directly used in heterogeneous information networks. It is very meaningful and challenging to study anomaly detection in heterogeneous information networks. In this paper, we introduce a framework which can detect anomaly in heterogeneous information network. We design the Relationship Prediction Neural Network Model (RPNN) to predict the relationship between nodes to learn the representation vector. Then the representation vector of the node type that we focus on is applied to the anomaly detection algorithm for detection. The experiments conducted on two real-world datasets show that our proposed model is effective compared with the state-of-the-art methods.