Ethereum transaction tracking : Inferring evolution of transaction networks via link prediction

Blockchain is an emerging technology which has attracted wide attention in recent years. As one of the blockchain applications, cryptocurrency has developed rapidly in recent years, attracting criminals to commit fraud and money laundering. Therefore, to better protect the legitimate interests of users and help formulate an effective supervision, it is necessary to track and follow transaction records on blockchain-based systems. This paper studies the problem of transaction tracking in Ethereum from a network perspective, aiming to study explainable strategies for money flow generation. We first collect the space-intensive transaction data from Ethereum blockchain and model them as temporal weighted multi-digraphs. A variety of tracking strategies considering different transaction factors (i.e., frequency and amount) are proposed, and the corresponding random-walk based link predictions method are designed for evaluation. Our method gets explainable results from the experiments, demonstrating that both transaction frequency and amount influence the generation of new transactions in Ethereum. This means when tracking the money flow among Ethereum accounts, we should pay more attention to those transaction paths having a shorter time interval and a larger amount. From these transaction features, the proposed random-walk based link prediction framework is found to be an effective method for transaction tracking. Furthermore, we show an application of transaction tracking via link prediction effectively enhance the ability to detect the suspicious accounts in Ethereum.