VulMDS: Precise Clone Vulnerability Detection via Multi-Dimensional Slicing
Shirun Liu, Aimin Yu, Zhengkai Qin · 2025
Detecting and locating clone vulnerabilities in source code is crucial for ensuring software security, maintenance, and evolution. Although numerous detection methods and tools are available, traditional approaches often suffer from low accuracy due to the lack of vulnerability-focused slicing, which introduces noise, and the challenge of distinguishing between similar patched and vulnerable code. In this paper, we introduce VulMDS, a precise method for clone vulnerability detection in source code. VulMDS employs multi-dimensional slicing to precisely extract only the lines related to vulnerable and patched changes, thereby effectively filtering out irrelevant noise. By generating vulnerability, patch, and context features to create a unique vulnerability finger-print, VulMDS effectively addresses the challenge of distinguishing between similar vulnerable and patched code. Experimental results demonstrate that VulMDS achieves a precision of 91% and a recall of 85%, significantly outperforming state-of-the-art methods. These results highlight VulMDS's potential to significantly improve clone vulnerability detection in software development.