Design And Implementation Of Pyramid Selling Organization Mining System In Financial Transactin Network

The criminal activities of pyramid selling are characterized by the complexity of the subjects involved,the large number of people involved,and the large amount of money involved,which poses a threat to the vital interests of the people,social stability and national security that cannot be ignored.Over the years,pyramid selling criminal activities have been repeatedly prohibited.With the rapid growth of financial transaction data,the traditional pyramid selling investigation methods that rely on manual screening and analysis face the problem of low efficiency.Thus,it is necessary to study efficient pyramid selling organization mining methods to assist economic investigators in pyramid selling investigation.This paper is oriented to financial transaction networks,and the research is conducted in three aspects:pyramid selling account detection,pyramid selling organization discovery,and pyramid selling organization role mining,and the main work is as follows.Firstly,a detection method based on temporal neural network is proposed for the problem of pyramid selling account detection.The method models the local and global information of transaction sequences,which can effectively solve the problem of discrete and hidden transaction sequence feature extraction,and is finally used to identify pyramid selling accounts.It can effectively improve the detection efficiency in the actual work of pyramid selling account detection.Secondly,the method based on local optimization is proposed for the problem of pyramid selling organization discovery.The method is oriented to the time-series transaction network,and identifies core nodes by calculating the node persistence core degree,identifies significant transaction flows based on the ant colony model,and simulates the dynamic changes of funds in the network to find the significant transaction flows in the network.Finally,combining the above identification results,the pyramid scheme is discovered based on pruning and growth strategies.The method is able to discover most of the pyramid selling members and their transaction relationships with low error cost in a practical pyramid selling organization discovery task.Thirdly,a network representation learning model based on graph neural network is proposed for the pyramid selling role mining problem.The model is able to learn transaction feature vectors for nodes in a semantic information-rich financial transaction network.Meanwhile,it solves the transaction feature similarity learning problem for nodes in the network through a loss mechanism based on the comparison of nodes with the full graph.Finally,the role mining is realized in the vector space by combining the clustering method.The method performs role mining in a real pyramid selling transaction network,and can efficiently distinguish nodes with different roles.Finally,based on the above research,a pyramid selling organization mining system is designed and implemented.Tests performed on the system show that the system can effectively detect individual pyramid selling accounts in financial transaction networks,discover pyramid selling organizations,and analyze the roles of members in the organizations to assist pyramid selling detection.