【Author】
Shao, Wei; Li, Hang; Chen, Mengqi; Jia, Chunfu; Liu, Chunbo; Wang, Zhi
【Source】ALGORITHMS AND ARCHITECTURES FOR PARALLEL PROCESSING, ICA3PP 2018, PT IV
【Abstract】In Bitcoin user identification, an important challenge is to accurately link Bitcoin addresses to their owners. Previously, some heuristics based on transaction structural rules or observations were found and used for Bitcoin address clustering. In this paper, we propose a deep learning method to achieve address-user mapping. We define addresses by their transactional behaviors and seek concealed patterns and characteristics of users that can help us distinguish the owner of a certain address from millions of others. We propose a system that learns a mapping from address representations to a compact Euclidean space where distances directly correspond to a measure of address similarity. We train a deep neural network for address behavior embedding and optimization to finally obtain an address feature vector for each address. We identify owners of addresses through address verification, recognition and clustering, where the implementation relies directly on the distance between address feature vectors. We set up an address-user pairing dataset with extensive collections and careful sanitation. We tested our method using the dataset and proved its efficiency. In contrast to heuristic-based methods, our model shows great performance in Bitcoin user identification.
【Keywords】Bitcoin; Blockchain; Deep learning; Bitcoin privacy
【摘要】在比特币用户识别中，一个重要的挑战是准确地将比特币地址与其所有者联系起来。此前，基于交易结构规则或观察发现的一些启发式算法被用于比特币地址聚类。在本文中，我们提出了一种深度学习方法来实现地址-用户映射。我们通过地址的事务行为来定义地址，并寻找隐藏的用户模式和特征，这些模式和特征可以帮助我们将某个地址的所有者从数百万个其他地址中区分出来。我们提出一个系统，学习映射从地址表示到紧欧氏空间，其中距离直接对应于地址相似性的度量。训练深度神经网络进行地址行为嵌入和优化，最终得到每个地址的地址特征向量。我们通过地址验证、识别和聚类来识别地址的所有者，其实现直接依赖于地址特征向量之间的距离。我们建立了一个地址-用户配对数据集，具有广泛的收集和仔细的卫生条件。通过对数据集的测试，证明了该方法的有效性。与基于启发式的方法相比，我们的模型在比特币用户识别方面表现出了良好的性能。
评论