Contextual Language Model and Transfer Learning for Reentrancy Vulnerability Detection in Smart Contracts

Abstract

The proliferation of smart contracts on blockchain technology has led to several security vulnerabilities, causing significant financial losses and instability in the contract layer. Existing machine learning-based static analysis tools have limited detection accuracy, even for known vulnerabilities. In this study, we propose a novel deep learning-based model combined with attention mechanisms for identifying security vulnerabilities in smart contracts. Our experiments on two large datasets (SmartBugs Wild and Slither Audited Smart Contracts) demonstrate that our approach successfully achieves a 90% detection accuracy in identifying smart contract reentrancy attacks (e.g. performing better than other existing state-of-the-art deep learning-based approaches). In addition, this work also establishes the practical application of deep learning-based technology in smart contract reentrancy vulnerability detection, which can promote future research in this domain.