| The virtual currency uses the blockchain as the underlying technology,so it has the characteristics of decentralization,immutability,anonymity and free circulation.As a financial innovation tool,virtual currencies such as Bitcoin are widely used because they are different from traditional financial instruments.However,the characteristics of virtual currency also make it gradually become a tool for illegal crimes such as money laundering,fraud,tax evasion and terrorist financing,which affects the stable development of the financial market.Since the virtual currency transaction mode is different from the traditional currency transaction mode,the traditional financial abnormal transaction detection method cannot be well applied in the virtual currency transaction scenario.The abnormal transaction detection of virtual currency is still being explored,and the existing research mainly has two shortcomings.First,due to the large amount of transaction data and no entity authentication,it is difficult to obtain as many user tags as possible,and most of the research starts from traditional unsupervised learning.Second,most of the existing detections start from a macro perspective.Although they can perceive anomalies from the overall structure of the transaction network,transaction participants tend to pay attention to specific types of abnormal behaviors and locations where they occur.In response to the above problems,this paper,from the perspective of behavior,analyzes the behavioral characteristics of each abnormal behavior,and proposes corresponding detection methods for three common abnormal behaviors in virtual currency transactions: money laundering,Ponzi scheme and phishing fraud.1.By introducing the graph convolutional neural network method of time series information,the money-laundering behavior detection capability of the Bitcoin transaction graph structure with timestamps is improved.The graph convolutional neural network is used to learn the structural features of the bitcoin transaction network,and the time series information of the bitcoin transaction network is obtained through the long short-term memory neural network in the time series interval,which improves the generalization ability of the model.2.By integrating the innovative SMOTE algorithm and GRU method,the detection and identification of Ponzi scheme smart contracts in Ethereum is realized.For the Ponzi scheme existing in Ethereum transactions,select opcode characteristics and transaction account characteristics,and use the SMOTE algorithm to oversample the unbalanced data set to balance the data set,and then use the recurrent neural network GRU to solve the Ponzi scheme smart contract.Perform classification and identification to realize the detection of Ponzi schemes in the Ethereum trading network.3.A method based on graph attention network combined with support vector machine is constructed to realize the detection of phishing and fraudulent accounts in the virtual currency trading network.Considering accounts as nodes and transactions as edges,the graph attention network is used to obtain graph embedding according to the node characteristics of frequent phishing fraud transactions reflected in the transaction network graph,and finally SVM is used as a classifier to realize phishing accounts in the Ethereum transaction network.identification.Through experiments on the publicly labeled virtual currency transaction data set,it is proved that the ability of the above three models to detect abnormal behaviors is improved compared with the traditional detection methods.Finally,the proposed detection model is integrated,and a prototype system is designed and implemented to warn of abnormal behaviors in virtual currency transactions.In practical application scenarios,the detection model proposed in this paper can effectively discover abnormal behaviors in the virtual currency trading network,and provide a reference detection scheme for the scientific and technological supervision of virtual currency. |
PDF Full Text Request