关键词: 比特币, 混币, 交易地址, 残差图卷积网络, 自注意力机制, 多层感知器

Abstract: Bitcoin has received attention for its decentralized peertopeer anonymity nature, but its pseudoanonymity makes the transaction still traceable. In order to meet the higher requirements of users for privacy, mixedcoin transactions came into being. However, mixedcoin transactions make it more difficult to trace the funds of bitcoin, and at the same time, it also becomes an accomplice in assisting criminals to illegally launder money. In order to prevent and combat money laundering and other financial crimes, this paper proposes a graph neural networkbased address recognition method for bitcoin mixedcoin transactions. Firstly, a rich and representative labeled address dataset is constructed; secondly, a residual graph convolution network ResGCN is constructed by adding residual connections for graph feature learning and embedding, which overcomes the gradient decay problem of the traditional graph convolution network GCN with the increase of the number of graph convolution layers; and then the selfattention mechanism is combined with the multilayer perceptron MLP for graph classification; and finally output the binary classification results. The experimental results show that the method in this paper can accurately recognize the mixedcoin transaction addresses.

Key words: bitcoin, mixed coin, transaction address, ResGCN, selfattention, MLP