CmaGraph: A TriBlocks Anomaly Detection Method in Dynamic Graph Using Evolutionary Community Representation Learning.

Anomaly detection for dynamic graphs, with graphs changing over time, is essential in many real-world applications. Existing works did not consider the accurate community structures in a dynamic graph. This paper introduces CmaGraph, a TriBlocks framework using an innovative deep metric learning block to measure the distances between vertices within and between communities from an evolution community detection block. A one-class anomaly detection block can capture the dynamic graph’s anomalous edges after these two functional blocks. This method significantly enhances the capability to detect anomalous edges by reconstructing the distance between the evolutionary communities’ vertices. We demonstrate the implications on three real-world datasets and compare them with the state-of-the-art method.