A Software Vulnerabilities Detection Framework Based on Graph Neural Network with Graph-Based Feature Representations

<p dir="ltr">With the rapid growth of software development and increasing system complexity, the risk of undetected software flaws has become a significant concern. This research explores an improved approach to identifying such flaws by representing source code in structured visual forms, allowing patterns of vulnerabilities to be more easily recognized. The study finds that different flaw types are best identified through specific representations, leading to higher detection accuracy. These findings offer valuable insights for the development of advanced, precise tools that assist developers in ensuring software safety, ultimately contributing to more secure digital systems for society at large.</p>