VulDet: Smart Contract Vulnerability Detection Based on Graph Attention Networks

Abstract

Due to the rapid development of blockchain, security issues caused by smart contract vulnerabilities are receiving increasingly widespread attention. Unfortunately, traditional smart contract vulnerability detection methods rely heavily on expert knowledge and elaborate rules, while neural network-based vulnerability detection methods have not yet achieved satisfactory accuracy either. In this paper, we propose a novel vulnerability detection method for smart contracts called VULDET. We first construct a contract graph based on the structure of the smart contract source code and combine security domain knowledge to attach additional features to nodes in the graph that are closely associated with vulnerabilities to highlight key nodes, and finally use graph attention networks for contract vulnerability detection. We apply VULDET to reentrancy vulnerability as well as timestamp dependency vulnerability detection and conduct extensive experiments, and the results show that our approach has significant advantages over existing methods.