Blockchain cryptocurrency abnormal behavior detection based on improved graph convolutional neural networks

Abstract

As cryptocurrency is widely used in the financial field, detecting anomalous trading behavior of blockchain-based cryptocurrencies has become increasingly important. Researchers have utilized graph convolutional neural networks (GCN) for detecting anomalous cryptocurrency transactions. However, GCN fails to fully capture the spatial information correlation between neighboring nodes when processing graph data, which limits the utilization of structural features in the model. Therefore, the performance of GCN may be constrained when dealing with complex, high-dimensional graph data. In this paper, we propose a cosine similarity-based graph convolutional neural network for detecting anomalous cryptocurrency transactions. Compared to traditional GCN models, our method can better utilize both the network structure features and spatial information correlation, and it performs better in processing infinite-dimensional graph data. Experimental results show that our model can effectively detect anomalous cryptocurrency transactions, thereby improving the security and reliability of cryptocurrency, and it has good application prospects.