A Survey on Graph Encoder for Node Anomaly Detection

Money laundering is the practice through which criminals disguise the origins of illegally obtained funds, enabling them to integrate these funds into the legitimate financial system. The United Nations estimates that annually, between 2% and 5% of global GDP approximately $0.8 to $2 trillion is laundered worldwide. This staggering figure underscores the critical need for effective identification and enforcement of anti-money laundering (AML) measures. A variety of techniques have been proposed to detect money laundering, primarily through the analysis of transaction graphs that illustrate money transfers between bank accounts. These methods often focus on the structural and behavioral dynamics of dense subgraphs associated with these transactions. However, a significant limitation of many of these techniques is their failure to account for the common practice of high-volume fund flows that traverse multiple interconnected bank accounts in a chain-like manner. Moreover, many current AML approaches tend to either achieve lower detection accuracy or incur high computational costs, rendering them less effective and practical for real-world financial systems. Consequently, this results in only a small fraction of money laundering activities being successfully detected and prevented, highlighting a critical gap in the ability to combat this global issue effectively. Addressing these challenges is essential to enhance the reliability and efficiency of money laundering detection efforts.