Attention-Driven Topology Adaptive Graph Convolutional Networks for Bitcoin Abnormal Transaction Detection

With the rapid development of e-commerce and internet finance, the number of participants in the Bitcoin field has been continuously increasing. Bitcoin is favored for its decentralized and anonymous nature, but this also leads to the misuse by criminals for illegal activities, resulting in an increasing number of abnormal transactions. In the era of big data, the development of efficient abnormal transaction detection algorithms is particularly crucial for maintaining the security and stability of the Bitcoin network. Existing algorithms face issues with accuracy when dealing with the dynamic and high-dimensional characteristics of Bitcoin transactions. This paper presents the Attention-Driven Topology Adaptive Graph Convolutional Networks for Bitcoin anomaly Transaction detection, referred to as ATA-GCN, a model specifically designed to adapt to the analysis of Bitcoin transactions in a big data environment. Initially, we employ time series analysis for dynamic modeling of Bitcoin transaction data, dividing the transaction information into transaction subgraphs across various time steps. Subsequently, we design a graph convolutional layer with multiple convolutional kernels, based on the Topology Adaptive Graph Convolutional Network (TAGCN), and aggregate the convolution results through an attention mechanism to more deeply capture the high-dimensional features of Bitcoin transactions. Finally, the aggregated results are fed into the graph convolutional layer, and the final prediction results are outputted. Experimental results demonstrate that the ATA-GCN model achieved an F1 score of 89.4%, which is 10-30% higher than that of traditional machine learning methods and over 30 % higher than that of the benchmark graph convolutional network, confirming its highly reliable and superior performance in detecting abnormal Bitcoin transactions in a big data context.