(1.School of Cyberspace Security，Information Engineering University，Zhengzhou 450001，China； 2.China Institute of Marine Technology & Economy，Beijing 100081，China； 3.Henan Key Laboratory of Network Cryptography Technology，Zhengzhou 450001，China)
Abstract： Due to its characteristics of huge circulation market value, user volume and anonymity of accounts, blockchain transactions are frequently threatened by abnormal behaviours such as theft, Ponzi scheme and fraud. This paper proposed a network representation learning model DeepWalk-Ba as feature extraction method, taking bitcoin as an example, to learn the network structure and attributes of blockchain transactions, and excavate hidden information from the neighborhood structure of transactions as features. Then, 5 supervised and 1 unsupervised machine learning algorithms were used for anomaly detection. The experiment indicated that the supervised model random forest performed best, with a precision of 99.3% and recall value of 86.4%. The detection effect was better than detection models using the traditional feature extraction methods.
Key words : lockchain；anomaly detection；network representation learning；random walk；machine learning
