摘要
Abstract
[Research purpose]In recent years,emerging blockchain technology has become a new pathway for money laundering crimes,severely threatening the normal financial order.There is an urgent need for technical methods to identify node addresses with suspi-cious transaction behaviors.This article proposes a method for identifying money laundering transactions in blockchain,aiming to establish a defensive barrier from a technical perspective.[Research method]This study designs the SWHA graph representation learning model,which establishes a mining function to capture the patterns of fund transfers between node addresses from the perspective of fund movement on the Tron blockchain.Additionally,it introduces a temporal prediction loss based on the Hawkes process that includes a fund purity term,enabling the model to learn integrated embedding vector representations of node addresses associated with money laundering behav-iors on the Tron network.The data are structured by using a meta-path construction method,and the label information in the dataset is de-rived from node addresses related to money laundering in actual case scenarios.[Research result/conclusion]The SWHA model differs from traditional graph representation learning,which focuses solely on the topological information of graph structure data while neglecting the degree of association between node addresses resulting from fund transfers.Compared to traditional graph representation learning meth-ods,the SWHA model demonstrates superior performance in classification testing,achieving higher precision,recall,F1 score and AUC metrics,as well as better results in node representation visualization analysis.Experiments show that the SWHA model can more effective-ly extract feature information related to money laundering node addresses.关键词
区块链洗钱/霍克斯过程/资金纯度/节点地址亲密度/注意力机制/图表示学习Key words
blockchain money laundering/Hawkes process/fund purity/node address closeness/attention mechanism/graph representa-tion learning分类
信息技术与安全科学
