A Hybrid Vertex Outlier Detection Method Based on Distributed Representation and Local Outlier Factor

Outlier detection is a basic task in network analysis, which is useful in many applications such as intrusion detection, criminal investigation, and information filtering. In this paper we proposed a hybrid outlier detection methods in complex networks based on Vertex Distributed Representation and Local Outlier Factor, with the aim to find abnormal vertexes that are apart from the group or community in complex networks. The proposed outlier detection method based on Vertex Distributed Representation (VDR) and Local Outlier Factor (LOF) is named as VDR-LOF. VDR-LOF maps vertexes or edges into a density continuous real-valued space, and then uses LOF algorithm to detection the outliers. We conducted experiments on American College Football Network and Enron Email Network, visualized the original networks and its corresponding feature map in 2D space, then we found the vertex outliers in the network.