基于交易时间衰减的以太坊恶意地址检测方法

Ethereum malicious address detection method based on transaction time decay

提出Trans-TAN模型,用于以太坊上的交易流向图中关联恶意地址的检测任务,模型改进基于Transformer模型的自注意力机制,根据以太坊地址的交易特点并受到牛顿冷却定理的启发,引入随时间交易的时间间隔衰减因素,同时融合以太坊地址间的相似度因素和交易金额因素。基于以上三方面,通过牛顿冷却定理的常微分方程解形式构建的地址关联矩阵,从而改进原有的自注意力矩阵。实验证明,Trans-TAN模型能够有效捕捉以太坊交易流向图过程中恶意节点地址的特征,在测试集中精准率(Precision)、召回率(Recall)和F 1指标优于传统的检测模型。

This article proposes the Trans-TAN model for detecting malicious addresses associated with transaction flow graphs on Ethereum.The model improves the self attention mechanism based on the Transformer model.Inspired by Newton′s cooling theorem and based on the transaction characteristics of Ethereum addresses,the Trans-TAN model introduces the time decay factor of transaction intervals over time.At the same time,it integrates the similarity factor between Ethereum addresses and the transaction amount factor.Based on the above three factors,the address association matrix is constructed in the form of a solution to the ordinary differential equation of Newton′s cooling theorem,thereby improving the original self attention matrix.Experimental results show that the Trans-TAN model can effectively capture the characteristics of malicious node addresses in the Ethereum transaction flow graph process,and its accuracy in the test set(Pre)is improved.The precision,recall,and F 1 metrics are superior to traditional detection models.