DNEA: Dynamic Network Embedding Method for Anomaly Detection.

Network embedding is a basic method for dynamic network analysis. Diverse dynamic network embedding methods have emerged in recent years, however, most existing works regard all observed information as true information, ignoring the anomalous edges in the dynamic networks. When the observed information is mixed with anomalous edges, the learned node embeddings cannot depict precise network properties effectively. Therefore, network embedding that can identify anomalous edges is a promising research topic. Inspired by this, we propose a novel end-to-end dynamic network embedding method called Dynamic Network Embedding for Anomaly Detection (DNEA), which can learn the robust node embeddings based on the neighborhood information and community structure in the dynamic networks. DNEA captures the dynamic characteristics of the network to reconstruct the network topology structure based on Stochastic Block Model (SBM), and detects anomalous edges from the perspective of reconstruction probability. In addition, DNEA utilizes negative sampling to handle the challenge of scarce anomaly labels. Experimental results on real-world datasets demonstrate DNEA can outperform the state-of-the-arts.