A Heterogeneous Graph-based Fraudulent Community Detection System

Anting Zhang, Bin Wu, Yinsheng Li

Abstract: Financial frauds by organized communities are becoming popular while financial requirements are too complicated for individuals to be provided. In this paper, a heterogeneous transaction graph is constructed based on interrelated fraudulent community risk factors. A heterogeneous subgraph embedding method is designed to identify fraud patterns. The fraudulent communities can be identified with node selection and community expansion algorithms. Based on the proposed graph and method, a fraudulent community detection and risk visualization system has been developed. The system provides a comprehensive view, in which the business information, associated information, and risk visualization are provided to improve risk identification and understanding experience. A number of critical technical indicators have been designed, among which are integrity, intuitiveness, practicality and authenticity. The experiments at the technical indicators have been conducted and discussed.