Reentrancy Vulnerability Detection Based on Graph Convolutional Networks and Expert Patterns

Abstract

Smart contracts as one of the most successful applications of blockchain. It holds digital currency with huge economic value. During the rapid development of smart contracts, vulnerabilities in the contracts have caused huge financial losses to the blockchain. This has strengthened researchers’ focus on smart contract security vulnerability detection. In this paper, we explore a vulnerability detection deep neural network-based method on combining features in both contract source code and bytecode forms. We conduct extensive experiments on the Ethereum smart contract datasets for reentrancy vulnerability. The experiment demonstrates that our method achieves 87% accuracy and 78% f1-score. Another experiment shows that our method maintains a good detection performance even when the feature part is missing.