Design And Implementation Of Pyramd Scheme Detection System In Financial Tansaction Network

With the rapid development of network and digitalization of the financial system,the capital flows handled by financial transaction institutions can reach tens of millions of orders of magnitude every day.The hidden illegal transactions,such as pyramid scheme and money laundering,can not be stopped.It is a common difficulty for the relevant investigative departments to detect pyramid scheme activities efficiently from massive transaction data and to crack down on economic illegal activities in time.Traditional manual analysis and detection mode can no longer be applied to large data environment.It is important to build an intelligent pyramid scheme behavior detection mode with artificial intelligence technology.At present,the traditional detection methods of abnormal transaction behavior,such as pyramid scheme,are largely based on the rule system constructed manually,or in most cases rely on manual judgment,which consumes manpower and material resources.In the current large dataset environment,these labor-intensive detection methods are also faced with time-consuming,inefficient and other issues,it is difficult to deal with the large dataset.Aiming at the problems existing in the field of pyramid scheme behavior detection,this paper proposes a practical and efficient pyramid scheme behavior detection method combined with machine learning algorithm.Firstly,aiming at the problem of abnormal transaction account recognition,this paper proposes a time-series neural network model based on multi-local attention mechanism,and a method of constructing financial time series.This model can extract discrete and complex transaction behavior features in financial time series by using multilocal attention windows to realize abnormal account recognition task.In addition,the standard bidirectional Gated Recurrent Unit(GRU)neural network model and the traditional GRU neural network model based on attention mechanism are constructed.The experimental results in abnormal transaction account detection tasks show that the accuary of the proposed model is 90.5%,and the F1 value is 85.8%,which achieves the best results in all models.Secondly,aiming at the problem of abnormal transaction pattern discovery in pyramid scheme,this paper proposes a sensitive amount pattern mining algorithm.Combining the long-distance pattern mining algorithm and the sensitive amount pattern extraction algorithm,we can find the sensitive amount co-occurrence pattern and the sensitive amount set of pyramid scheme.The average test accuracy is over 90%.In addition,the accuracy of pyramid scheme accounts reaches 82% in the sensitive account set extracted according to the sensitive amount set in this paper.Thirdly,aiming at the problem of high-level node discovery in pyramid organizations,this paper proposes a high-level node discovery model.The model extracts sensitive account sets,builds account transaction feature vectors and clusters them to realize role discovery of high-level nodes in pyramid organizations.The experimental accuracy reaches 84.7%.Finally,based on the above research content,the pyramid behavior detection system is designed and implemented.It provides a practical and effective method for identifying pyramid account in financial transaction network,sensitive amount and high-level organization discovery.The test results prove that the system can achieve the expected design goals.