Money Laundering Detection Based on Suspicious Subgraph in Bitcoin Networks

Abstract

Money laundering in Bitcoin networks threatens social stability and undermines global economic security, making its detection a top priority for governments and relevant sectors. However, existing detection methods often struggle to extract critical transaction information due to the inherent heterogeneity and massive noise in Bitcoin networks, resulting in low detection performance. To address this challenge, we propose a novel suspicious subgraph-enhanced method to identify covert money laundering activities in Bitcoin networks. Our core innovation is a semantic-driven subgraph analysis strategy, realized as a proactive financial purification mechanism. This strategy fundamentally reframes the problem by focusing on semantic relevance over mere topology, thereby achieving active denoising and functional isolation. Crucially, it successfully mitigates the inherent feature dilution problem in Graph Neural Networks (GNNs) by providing a semantically pure, signal-rich subgraph input. Experiments on large-scale Bitcoin transaction data confirm the practical superiority of our approach, demonstrating a significant 20% increase in recall compared to traditional Graph Convolutional Network (GCN) models, with high overall accuracy (96%) and recall (98%), thus meeting stringent regulatory requirements.