Abstract: Blockchain offers users anonymity and facilitates the decentralized transfer of value. However, malicious attackers might employ phishing or other fraudulent methods to steal assets and withdraw them from cryptocurrency exchanges by designing complex transaction interactions. In this paper, we address this challenge by presenting a valuedriven transaction tracking and ranking method tailored for Ethereum. In this approach, we collect a transaction dataset of up to 27 GB from 12 Ethereum attack cases with fraud amounts exceeding one million US dollars, and construct an address graph to describe the relationship between addresses. We then invoke token liquidity pool data from the onchain data to represent the historical price of assets and determine the weight coefficients for transactions in the graph. Finally, we introduce a dynamic residual scaling mechanism based on value proportion to optimize the address graph structure by optimal value flow paths. Experimental results show that the proposed method achieves a recall rate of 89.24%, which represents a notable improvement of 7%, 20%, and 37% over transaction tracing rank (TTR), APPR, and Haircut algorithms, respectively, confirming the effectiveness of the proposed method.

Key words: blockchain, transaction tracing, PageRank, token value, fraudulent accounts