Comparative Analysis of Graph Representation Techniques for Code Vulnerability Detection

Comparative Analysis of Graph Representation Techniques for Code Vulnerability Detection

연구 분야: Strategies

학회: 2023 International Conference on Engineering and Emerging Technologies (ICEET)

The surge in software development and products that use software have intensified the risk of software vulnerabilities. The difficulty of maintaining code security has been exacerbated by the growing software ecosystem and complexity of contemporary software systems. Code vulnerability detection is important in preventing these vulnerabilities, but current methods are time-consuming and may overlook certain vulnerabilities. Deep learning, specifically tree/graph-based neural network, may be a promising solution as it preserves hierarchical and structural information in code, leading to better performance in vulnerability detection. This research engages in a more comprehensive investigation in discussing different methods of code representation, including Abstract Syntax Trees (AST), Control Flow Graphs (CFG) and Program Dependence Graphs (PDG). These graphs undergo thorough analysis and are examined to pinpoint the critical information they extract and how it contributes to vulnerability detection. Notably, it is discovered that distinct graph representations excel at identifying different types of vulnerabilities. The information captured from each graph representations can be used to identify the type of vulnerability and differentiate vulnerable code from non-vulnerable code. Such finding lays the foundation for a more advanced software vulnerability detection tool based on graph representation of codes, offering software vulnerabilities can be efficiently addressed.