| With blockchain technology as the underlying backing,Ethereum has become an excellent example of blockchain 2.0 development,which has gained great development and support in the economic and financial fields while also having various issues.The anonymity of blockchain accounts leads to the inaccessibility of users’ attribute information,and criminals try to inject funds into the Ethereum trading system to evade regulation and engage in phishing fraud,money laundering and other illegal transactions,which poses a serious threat to the security and stability of Ethereum.At present,there are relatively few studies at home and abroad on the detection of abnormal transactions in the Ethereum blockchain,and the existing detection methods have problems such as complicated account feature extraction and high rate of abnormal transaction miss detection.Therefore,this thesis focuses on two aspects of feature mining of anonymous accounts and abnormal transaction behavior detection in the Ethereum transfer transaction network to carry out research on the identification of abnormal transactions in Ethereum based on graph convolutional network,to characterize the attribute information of anonymous accounts and detect abnormal transaction behaviors caused by phishing fraud in Ethereum,which provides a new perspective to ensure the security and stable development of Ethereum.The main research work of this thesis is as follows:(1)A time-amount multi-channel walk based embedding method is proposed,aiming at the anonymity problem of Ethereum transaction accounts.It uses breadthfirst traversal-based and depth-first traversal-based random walk generators to capture the features of anonymous accounts regarding transaction amount-time connection pattern multiplicity simultaneously in the temporal amount multiplicity graph through a multi-channel sampling strategy,which characterizes the portrait of anonymous accounts and is used to build an Ethereum transaction property graph based on the features of anonymous accounts.The proposed method is proved to be effective in characterizing the information of anonymous accounts through visualization and link prediction experiments,with improved results compared to similar methods.(2)A graph convolutional network based method for detecting ethereum anomalous transactions is proposed.It extracts a comprehensive feature representation of the network structure and attribute features of the Ethereum transaction attribute graph by a classifier built on the graph convolutional network.When extracting the structural features,an attention mechanism is applied to capture the feature differences between nodes and neighboring nodes to represent node features.Finally,the learned integrated feature distribution is mapped to the sample label space using the fully connected network to complete the training and testing of the model for identifying the abnormal transactions triggered by phishing fraud accounts.The method can better solve the classification problem of transaction behavior recognition oriented to ethereum verified by ablation experiments and classification test experiments.(3)A prototype system for Ethereum abnormal transaction detection is designed and implemented.The system integrates the Ethereum abnormal transaction detection model proposed in this thesis,so as to verify the use effect of the model in actual application scenarios,thus providing more information for the supervision of Ethereum transactions.possibility.Identify phishing transactions triggered by detected phishing fraud accounts. |
