Scalable Graph-Based Detection of Fraud Rings in Large-Scale Networks

Graphs are widely used to model relational data in various domains, including social media, e-commerce, telecommunications, and finance. Graph analytics is one of the most popular techniques for analyzing connectivity patterns in communication networks and identifying suspicious behaviors. However, detecting fraud rings at scale is significantly challenging since the volume of graph data is growing exponentially. Moreover, many Python-based graph libraries rely on in-memory computation, which often struggles with large-scale networks. To address these limitations, we extend our previous semiring-based graph computation framework into a domain-specific solution for large-scale fraud ring detection. We formalize key graph patterns that indicate the presence of fraud rings and develop a general detection algorithm based on semiring operations. Our algebraic approach operates efficiently on a hybrid architecture that can scale beyond RAM constraints. Furthermore, it provides interpretable results, ensuring mathematical transparency to meet regulatory demands for explainability. Our approach is developed in C++, and it can be easily called in Python. An Experimental comparison with state-of-the-art Python packages shows that our approach has comparative performance for both small and large graphs.