基于多任务LSTM+TGCN的虚拟货币反洗钱检测研究  

Research on Anti-Money Laundering Detection of Virtual Money Based on Multi Task LSTM+TGCN

作　　者：陈宇琪 周高远 徐铭鑫 钱汉伟 Chen Yuqi;Zhou Gaoyuan;Xu Mingxin;Qian Hanwei(Department of Computer Information and Cybersecurity,Jiangsu Police Institute,Nanjing 210031,China;State Key Laboratory for Novel Software Technology,Nanjing University,Nanjing 210093,China)

机构地区：[1]江苏警官学院计算机信息与网络安全系,南京210031 [2]南京大学计算机软件新技术国家重点实验室,南京210093

出　　处：《信息化研究》2024年第6期14-21,共8页INFORMATIZATION RESEARCH

基　　金：江苏高校哲学社会科学研究项目(No.2024SJYB0344);国家自然科学基金项目(No.72401110)。

摘　　要：随着虚拟货币在洗钱等非法金融活动中的日益广泛应用,传统的反洗钱检测技术面临新的挑战。本文深入探讨了虚拟货币洗钱的技术机制,并提出一种改进的检测方法。该方法通过自注意力机制和时间注意力机制,增强了模型对潜在欺诈特征的识别能力。利用长短期记忆网络(LSTM)动态图学习器捕捉金融网络的动态变化,并通过时空图卷积网络(TGCN)有效处理交易数据的时空信息。我们提出的多任务TGCN模型不仅同时进行交易金额预测和异常检测,还采用焦点损失函数来解决类别不平衡问题,并通过多任务学习框架平衡两个任务的损失。研究中还包括了关键的数据预处理步骤、模型训练、评估以及特征重要性分析流程。实验结果表明,LSTM+TGCN模型在识别洗钱交易方面表现出较高的准确率,证明了其在反洗钱领域的有效性。With the increasingly extensive application of virtual currency in illegal financial activities such as money laundering,the traditional anti money laundering detection technology is facing new challenges.This research deeply discusses the technical mechanism of virtual money laundering,and proposes an improved detection method.This method enhances the model's ability to identify potential fraud features through self-attention mechanism and time attention mechanism.The Long Short-Term Memory(LTSM)dynamic graph learner is used to capture the dynamic changes of the financial network,and the spatiotemporal information of the transaction data is effectively processed through the Temporal Graph Convolution Network(TGCN).The multi task TGCN model proposed by us not only forecasts the transaction amount and detects anomalies simultaneously,but also uses the focus loss function to solve the category imbalance problem,and balances the losses of the two tasks through the multi task learning framework.The research also includes key data preprocessing steps,model training,evaluation and feature importance analysis processes.The experimental results show that the LSTM+TGCN model shows high accuracy in identifying money laundering transactions,which proves its effectiveness in the field of anti-money-laundering.

关 键 词：长短期记忆网络 时空图卷积网络 虚拟货币 反洗钱 

分 类 号：TP393.08[自动化与计算机技术—计算机应用技术]