Clue2Group AI Framework Boosts Money Laundering Detection

This paper introduces a novel approach to Money Laundering Group Discovery (MLGD) that better aligns with real-world Anti-money-laundering (AML) investigations. Current methods often produce only node-level risk alerts or passively search entire networks, which doesn't match how analysts typically start with a specific clue and expand their investigation. The proposed Clue-Guided Group Discovery (CGGD) framework aims to progressively recover complete laundering group structures from an initial set of clues through analyst interaction. The core of this framework is Clue2Group, which first builds a compact local investigation context to minimize noise and preserve critical chain-like and cycle-like laundering patterns. It then uses a multi-semantic local-temporal Graph Neural Network (GNN) to estimate a clue-conditioned local risk field. Finally, Clue2Group integrates risk, structural, and prior-pattern evidence to reconstruct a coherent money laundering group. Experiments on two large-scale AML benchmarks demonstrate that Clue2Group provides a practical, clue-driven analysis framework, bridging the gap between graph-based AML research and actual investigation workflows.