A Dual Convolution Hybrid Multi-Model for Code Vulnerability Detection

Existing vulnerability detection models can typically detect only a limited range of vulnerabilities. As the variety of vulnerabilities increases, performance will be reduced. Graph Convolutional Networks (GCNs) frequently encounter challenges with capturing long-distance dependencies, which impacts their detection effectiveness. To address these limitations, this paper proposes a dual-convolution model to detect code vulnerability. Specifically, the proposed model utilizes subgraphs created through program slicing and integrates GCN with dual-convolution layers to effectively capture both global and local code structural information. Extensive experiments on datasets demonstrate the accuracy improvement of our model.