Clue2Group framework lets analysts uncover money laundering rings from initial clues

A new paper from arXiv by researchers Boyang Wang and Jianing Cao tackles the mismatch between existing graph-based anti-money laundering (AML) methods and real-world investigation workflows. Current approaches either produce node-level risk alerts (e.g., suspicious accounts) or blindly search the entire network for hidden groups. Neither aligns with how AML analysts actually work: they start with a concrete clue—like a suspicious transaction or a known bad actor—and gradually expand to uncover the full criminal organization. To bridge this gap, the authors propose Clue-Guided Group Discovery (CGGD) and introduce Clue2Group, a framework that operationalizes this interactive, clue-driven process.

Clue2Group first constructs a compact local investigation context around the initial clue, reducing noise while preserving critical chain-like and cycle-like laundering structures. It then estimates a clue-conditioned local risk field using a multi-semantic local-temporal graph neural network (GNN), capturing both structural and temporal patterns in financial networks. Finally, it integrates risk, structural, and prior-pattern evidence to recover a coherent money laundering group. The framework was tested on two large-scale AML benchmarks and showed strong performance, enabling practical clue-driven analysis that aligns with real investigative needs. This work represents a significant step toward making graph-based AML research more directly applicable in operational settings.