Reentrancy Vulnerability Detection Based on Semantic-Aware Graph Convolutional Neural Network

Aiming at the problems of low accuracy of conventional smart contract vulnerability detection methods, as well as the single form of feature extraction of deep learning-based methods, which are unable to adequately represent the rich semantic and structural information present in the source code, we propose a smart contract reentrancy vulnerability detection method based on semantic-aware graph convolutional neural network. The method first constructs the smart contract graph, extracts the graph structure features in the smart contract source code by the degree-free graph convolutional neural network, then slices the contract source code, extracts the sequence features by using BiGRU and attention mechanism, and finally splices the extracted graph structure features and the sequence features to find reentrancy vulnerability. The experimental findings demonstrate that, when compared to deep learning and conventional detection approaches, the suggested method in this work exhibits improved accuracy, recall, precision, and Fl score, with a detection accuracy of 89.77%.