Embedded Files
RQ1: Can SnapVuln outperform state-of-the-art baselines?
RQ1: Can SnapVuln outperform state-of-the-art baselines?
Answer to RQ1: The overall performance of SnapVuln outperforms existing approaches on the real-world dataset D2A and the synthetic dataset Juliet.
RQ2: Can the completeness of the extracted vulnerability semantics affect the performance in vulnerability detection?
RQ2: Can the completeness of the extracted vulnerability semantics affect the performance in vulnerability detection?
Answer to RQ2: The completeness of the extracted vulnerability semantics affects the performance of vulnerability detection. Although some vulnerability-irrelevant semantic information is introduced at the same time, it can still improve detection performance. It also indicates that precise slicing criteria are demanded to reduce the noise and further improve the performance.
RQ3: Can the precision of the extracted vulnerability semantics affect the performance on vulnerability detection?
RQ3: Can the precision of the extracted vulnerability semantics affect the performance on vulnerability detection?
Answer to RQ3: The precision of the extracted vulnerability semantics is vital to the performance of vulnerability detection. Compared with other slicing approaches, we confirm that our slicing algorithms extract more precise vulnerability semantics.
RQ4: How does the number of subgraphs extracted by the proposed slicing algorithms affect the performance of vulnerability detection?
RQ4: How does the number of subgraphs extracted by the proposed slicing algorithms affect the performance of vulnerability detection?
Answer to RQ4: SnapVuln achieves best performance with k set to 16 on D2A and 6 on Juliet respectively.
Page updated
Google Sites
Report abuse