基于机器学习的比特币去匿名化方法研究

比特币是一种去中心化的匿名加密货币，是目前使用最广泛的数字资产之一，具有匿名性、无主权、无地域限制的特点，匿名性的特性也使得比特币被广泛应用于各种犯罪活动。为实现比特币的去匿名化，提出一种联合特征构造方法并构建随机森林与Softmax相结合的分类模型。为更好地区分不同类型比特币的交易行为，引用交易实体的概念，按照联合特征构造方法分别从地址、实体与交易网络结构3个方面在海量的交易数据中构造特征，并将其整合成联合特征向量。实验结果表明，该实体分类模型的类别识别精确率超过0.92，其能够有效提升执法机构对虚拟货币犯罪行为的调查取证能力。

Research of De-Anonymizing Method Based on Machine Learning for Bitcoin

Bitcoin is one of the most widely used digital assets.It is a kind of decentralized anonymous cryptocurrency, and has no sovereignty or geographical restrictions.However, its anonymity makes it abused in illegal activities.To realize de-anonymization for Bitcoin, a classification model combining Random Forest(RF) and Softmax is proposed along with a joint features construction method.To distinguish the transactions of different kinds of Bitcoin, the concept of entity is introduced.On this basis, the joint feature construction method is used to construct features in massive transaction data from the aspects of address, entity and transaction network, and the constructed features are integrated into a joint feature vector.The experimental results show that the proposed entity classification model displays a classification accuracy of over 0.92.It can effectively assist law-enforcement agencies in investigating and collecting evidence of illegal activities involving virtual currency.