TS-GGNN: Combining Graph and Sequence Features for Vulnerability Detection in Source Code

Software vulnerability detection is crucial for maintaining the security and stability of software systems. In this paper, we propose a novel neural network model called TS-GGNN to address the problem of vulnerability detection in source code slices. The TS-GGNN model effectively captures both local and global features of vulnerable code by fusing sequence features with graph features. To achieve this, we utilize graph structure and sequence structure learning approaches to comprehensively extract valuable information from the source code slices. Our experiments are conducted on the SARD dataset, which consists of 61,638 code samples annotated for the presence or absence of vulnerabilities. The results demonstrate that TS-GGNN has the best vulnerability detection performance, with an accuracy of 99.4%, a precision of 98.81%, and an F1 score as high as 99.4% thereby validating the effectiveness of the TS-GGNN model in capturing features relevant to software vulnerabilities.