Anomaly detection by discovering bipartite structure on complex networks

Abstract Anomaly detection is a classic problem on complex networks. An anomaly detection method based on network projection is proposed in this study on networks with fundamental bipartite connection relationships and repeated interactions, such as the Internet and computer networks. First of all, two network partition algorithms are advanced to discover the bipartite structure of the network. Then, put forward a similarity metric based on the cosine similarity function to construct the projection network. Finally, the metrics of vertices in the projection network are used as input of the one-class SVM algorithm to detect anomalies. Experiments on simulation datasets and real datasets illustrate that our method achieves higher precision in identifying anomalous addresses than traditional approaches for large-scale Internet and computer networks with the bipartite structure.