Ethereum Account Classification Based on Graph Convolutional Network
Tao Huang, Dan Lin, Jiajing Wu
Abstract
Accounts in Ethereum are found to be involved in various services or businesses. Account classification can help us detect illegal behavior, track transactions, and de-anonymize the Ethereum transaction system. In this brief, we make use of Graph Convolutional Network (GCN) to solve the account classification problem in Ethereum. We model the Ethereum transaction records as a large-scale transaction network and find that the network is with high heterophily, in which accounts with different features and different labels are connected. In order to solve this problem, we propose a GCN-based model called EH-GCN. The experimental results on a realistic Ethereum dataset show that the proposed method achieves the most advanced classification performance, and results on benchmarks show it produces a competitive performance under homophily.