Network Embedding Analysis for Anti-Money Laundering Detection

2025-09-12

AI Summary

Money launderers deliberately engineer “anti-central” nodes to evade conventional graph centrality–based detection methods. Method: This paper proposes a risk identification framework integrating network embedding and structural features on directed financial transaction graphs. We employ node2vec to learn low-dimensional node representations and introduce a novel metric—*spread number*—to quantify each node’s diffusion capacity across multi-hop paths. A composite risk score *R* is then constructed by jointly leveraging degree centrality, betweenness centrality, and embedding-based similarity. Contribution/Results: The method effectively identifies structurally critical yet anomalously low-centrality accounts—common hallmarks of obfuscated money laundering activity. Evaluated on real-world banking data, it precisely pinpoints high-*R* clusters corresponding to latent money laundering syndicates. It significantly enhances detection sensitivity for sophisticated laundering patterns, including circular fund transfers and hierarchical, nested transaction structures.

Technology Category

• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Machine Learning: Graph-based Machine Learning
• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Web Mining and Content Analysis: Bridging structured and unstructured data
• Responsible Web: Measurement, analysis, and circumvention of Web censorship

Abstract

We employ network embedding to detect money laundering in financial transaction networks. Using real anonymized banking data, we model over one million accounts as a directed graph and use it to refine previously detected suspicious cycles with node2vec embeddings, creating a new network parameter, the spread number. Combined with more traditional centrality measures, these define an aggregate score $R$ that highlights so-called anti-central nodes: accounts that are structurally important yet organized to avoid detection. Our results show only a small subset of cycles attain high $R$ values, flagging concentrated groups of suspicious accounts. Our approach demonstrates the potential of embedding-based network analysis to expose laundering strategies that evade traditional graph centrality measures.

Problem

• Detecting money laundering in financial networks
• Identifying structurally important but evasive accounts
• Combining embedding and centrality for improved detection

Innovation

• Uses node2vec embeddings on transaction networks
• Introduces spread number as new network parameter
• Combines embeddings with centrality for anti-central detection