Ethereum Smart Contract Account Classification and Transaction Prediction Using the Graph Attention Network

Abstract

This study explores the application of a Graph Attention Networks version 2 (GATv2) model in analyzing the Ethereum blockchain network, addressing the challenge posed by its inherent anonymity. We constructed a heterogeneous graph representation of the network to categorize contract accounts (CAs) into different decentralized application (DApp) categories, such as DeFi, gaming, and NFT markets, using transaction history data. Additionally, we developed a link prediction model to forecast transactions between externally owned accounts (EOAs) and CAs. Our results demonstrated the effectiveness of the heterogeneous graph model in improving node embedding expressiveness and enhancing transaction prediction accuracy. The study offers practical tools for analyzing DApp flows within the Web3 ecosystem, facilitating the automatic prediction of CA service categories and identifying active DApp usage. While currently focused on the Ethereum network, future research could expand to include layer 2 networks like Arbitrum One, Optimism, and Polygon, thereby broadening the scope of analysis in the evolving blockchain landscape.