一种基于特征增强的图神经网络比特币非法交易检测方法OA

Detection of Bitcoin illegal transactions is a significant challenge in blockchain technology,particularly when faced with complex transaction patterns and issues of class imbalance.This paper proposes a feature-enhanced graph neural network approach aimed at improving the accuracy of Bitcoin illegal transaction detection.First,the BERT model is employed to enhance transaction features,leveraging its powerful contextual modeling capabilities to extract more expressive features.Second,an innovative model architecture is designed,combining LSTM with a dual-channel ONGNNConv,where the latter passes its output to an attention mechanism for weighted aggregation,thereby enabling more effective capture of latent patterns in the transaction network.Finally,to address the class imbalance problem,a weighted binary cross-entropy loss is incorporated into the loss function to enhance the detection of illegal transactions.Experimental results demonstrate that the proposed method outperforms existing baseline models across multiple evaluation metrics,validating its effectiveness and robustness in Bitcoin illegal transaction detection.