A Graph Neural Network-Based Smart Contract Vulnerability Detection Method with Artificial Rule

As blockchain technology advances, the security of smart contracts has become increasingly crucial. However, most of smart contract vulnerability detection tools available on the market currently rely on artificial-predefined vulnerability rules, which result in suboptimal generalization ability and detection accuracy. Deep learning-based methods usually treat smart contracts as token sequences, which limit the utilization of structural information and the integration of artificial rules. To mitigate these issues, we propose a novel smart contract vulnerability detection method. First, we propose an approach for constructing contract graph to capture vital structural information, such as control- and data- flow. Then, we employ a Wide & Deep learning model to integrate the structural feature, sequencial feature, and artificial rules for smart contract vulnerability detection. Extensive experiments show that the proposed method performs exceptionally well in detecting four different types of vulnerabilities. The results demonstrate that integrating structural information and artificial rules can significantly improve the effectiveness of smart contract vulnerability detection.