Ethereum, the most popular public blockchain with the capability of smart contracts and the cryptocurrency Ether, is escalating in the number of account addresses and transactions since its birth. Due to the decentralisation of the Ethereum blockchain and the anonymity of its users, Ethereum serves as a noteworthy environment for malicious activities that are difficult to unearth. As a result, understanding the behaviors of the account addresses on Ethereum has become an imperative problem receiving much attention very recently. Existing works for such task mainly rely on extracting statistical features of account addresses and applying machine learning techniques to group or identify them. However, seldom prevailing approaches take temporal information and high-order interactions among the account addresses into consideration. To this end, we propose a novel approach coined THCD (T emporal H igh-order proximity aware C ommunity D etection) for behavior analysis on Ethereum from the perspective of graph mining. First, frequent temporal motifs are mined over a transaction graph constructed by the Ethereum block transactions. Next, we define the high-order proximity between two accounts based on these temporal motif occurrences. Finally, a novel temporal motif-aware community detection method is devised to find account communities over the defined high-order proximity. Experiments on four real datasets constructed from Ethereum blocks demonstrate the effectiveness of our approach. Some discovered suspicious accounts are confirmed by real-world reports. Meanwhile, THCD is scalable to large-scale transaction datasets.

以太坊（Ethereum）是最流行的具有智能合约和加密货币以太坊功能的公共区块链，自诞生以来，其账户地址和交易数量正在逐步增加。由于以太坊区块链的去中心化及其用户的匿名性，以太坊成为难以发掘的恶意活动的重要环境。结果，最近在以太坊上了解账户地址的行为已经成为当务之急。用于该任务的现有作品主要依赖于提取账户地址的统计特征并应用机器学习技术来对它们进行分组或识别。但是，很少有主流方法将时间信息和帐户地址之间的高级交互考虑在内。为此，Ť emporal ħ IGH-顺序接近感知Ç ommunity d etection），用于从图挖掘的角度上复仇行为分析。首先，在以太坊区块交易构建的交易图上挖掘频繁的时间主题。接下来，我们根据这些时间主题出现来定义两个帐户之间的高阶接近度。最后，设计了一种新颖的可感知时间主题的社区检测方法，以查找定义的高阶邻近度上的帐户社区。在以太坊区块构建的四个真实数据集上的实验证明了我们方法的有效性。真实世界的报告证实了一些发现的可疑帐户。同时，THCD可扩展到大规模交易数据集。