Telecomm Fraud Detection via Attributed Bipartite Network

Internet malicious incident occur frequently in recent years, bringing huge loss in different fields such as telecommunication, finance, etc. Such incidents generally entail fraudulent behaviors, which deviate from normal behavioral patterns. Particularly, in telecommunication, those anomalous interactions can be manifested by the network formed through interactions between callers and callees. Telecommunication network is indeed a bipartite, directed, and attributed network, and the anomalous network structure can be better captured by simultaneously considering these distinctive perspectives. To that end, in this paper, we propose a novel metric to measure the abnormality of dense subgraph structure by considering the structure, temporal changes, as well the attributes on nodes and edges. We further formulate an objective function and propose a greedy approach to discover the structure and corresponding fraudsters. The proposed metric and algorithm are experimented on a real-world telecommunication network dataset, which is shown to achieve competitive fraud detection performance than the baseline methods.