Dataset for "A Comparative Study of GNN-based and LLM Techniques for Detecting Software Vulnerabilities"

A Comparative Study of GNN-based and LLM Techniques for Detecting Software Vulnerabilities datasets Contains test and train splits from 3 different datasets: BigVul MegaVul ProjectKB results Contains merged results from the GNN-based, the LLM-based, the combined methods, and the manual evaluation: line level results for all test records and all methods for manually evaluated records and methods aggregated results for all test records and all methods per cwe for selected methods barcharts to visualize manual evalutation results Interpretation of ground_truth and method columns in line level results: 0 means secure 1 means vulnerable