Deep Learning-Based Smart Contract Vulnerability Detection Model

Abstract

As blockchain technology matures, the application scope of smart contracts continues to expand. Security issues in smart contracts have become an important research focus. In view of the limitations of traditional vulnerability detection methods such as limited vulnerability types and high false positive rate, this paper proposes a smart contract vulnerability detection model based on deep learning. By constructing token sequence sets and data flow graphs and inputting them into the pre-training model of this article, we obtained a pre-training model optimized for Solidity code and implemented a multi-label classifier in the downstream detection layer. The proposed model can detect integer overflow vulnerabilities, transaction sequence dependency vulnerabilities, timestamp vulnerabilities, and reentrancy vulnerabilities in the target contract. Experiments show that this model is better than other detection tools in detection effect.