Research And System Implement Of Anti-Money Laundering Methods Based On Graph Embedding Technology

Money laundering activities pose serious harm to the development of the economy and society.However,existing graph embedding techniques have the following problems in capturing the characteristics of money laundering activities: firstly,existing graph embedding techniques can only capture the relationship features at the node level in graph structure data,but cannot effectively capture the relationship features at the relationship level.In addition,the flow of funds between accounts is essentially a directed graph,where nodes on any path in the graph have sequence features.Existing graph embedding techniques do not take into account the sequence features in the path,thus unable to effectively capture the flow trajectory characteristics of funds.In response to the above issues,the main work of this article is as follows:Firstly,aiming at the problem that the existing graph embedding technology cannot effectively capture the relationship characteristics at the relationship level,a Multi-Entity Relation Embedding model(MERE)based on random walk is proposed.The innovation of this model is that it extracts several transaction chains from the graph with relationship characteristics at the relationship level through random walk,and embeds these transaction chains into the node vector.The node embedding vector obtained through this method not only includes the relationship features at the node level,but also the relationship features at the relationship level.The node classification experiments on the simulation dataset show that the MERE model outperforms existing graph embedding techniques in terms of accuracy,precision,recall,and f1 score.Secondly,a Fund Flow Trajectory Embedding(FFTE)model based on recurrent neural networks is proposed to address the issue of existing graph embedding techniques not considering the sequence characteristics of nodes on any path in the graph.The innovation of this model lies in its node aggregation method,which introduces the concept of hidden layers in recurrent neural networks based on graph convolutional networks,and uses the state of the hidden layers as the final embedding vector of nodes.The node embedding vector obtained through this method preserves the sequence features in any path in the graph,which can effectively capture the trajectory characteristics of fund flow.The experiments on the simulation dataset show that compared with existing graph embedding techniques,the proposed FFTE model has significant improvements in accuracy,accuracy,recall,and f1 score.Finally,based on the two models proposed above,a set of anti-money laundering system based on graph embedding technology was designed and implemented.From the perspective of system application,the system mainly includes functions such as system user management,transaction data management,anti-money laundering model management,anti-money laundering strategy management,real-time money laundering transaction monitoring,abnormal transaction analysis,etc.Based on transaction graphs,the system realizes visual analysis and monitoring of anti-money laundering.In conclusion,in view of the limitations of the existing graph embedding technology in capturing the relationship characteristics and capital flow trajectory characteristics at the relationship level in money laundering activities,this paper proposes a Multi-Entity Relation Embedding model based on random walk and a capital flow trajectory embedding model based on recurrent neural network to detect money laundering accounts,and designs and implements an anti-money laundering system based on graph embedding technology.After experimental and systematic verification,the two models proposed in this article are effective in capturing money laundering activities...