Source Code Vulnerability Detection Technology Based on Code Semantic Contribution
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
This research is dedicated to developing source code vulnerability detection technology based on code semantic contribution to address software security challenges. The research preprocesses source code functions by constructing a dictionary library of common programming naming conventions, builds data dependency graphs and control dependency graphs using the Joern tool, adopts BPE word segmentation technology, and combines sentence embedding variants and LSTM graph embedding for semantic feature extraction and embedding to achieve precise detection. The construction of the experimental environment takes into account both hardware performance and software compatibility. The Big-Vul, SARD and NVD datasets were selected and used for the experiment after cleaning and stratified sampling processing. Compared with methods such as Checkmarx, SySeVR based on BLSTM, VulCNN based on CNN, Devign based on graph neural networks, and LineVul based on Transformer, the results show that: The semantic contribution method of this study performed outstandingly in the comprehensive performance indicators, with a recall rate of 87.52%, an accuracy rate of 90.13%, a precision rate of 88.76%, and an F1 value of 0.881. It shows obvious advantages under different types of vulnerability detection, different code scales and training data volumes, and has lower false positive and false negative rates. This technology provides a more effective way for source code vulnerability detection and is expected to enhance the overall security of the software.
Fields of Study & Subjects
Software Engineering Research
Software Testing and Debugging Techniques
Web Application Security Vulnerabilities
Publication details
- Journal / source
- International Conference on Communication Networks and Machine Learning (CNML)
- Pages
- 371–376
- Publication date
- 30 January 2026
- Language
- EN