Testing Semantic Clone Detection Candidates

In this study, a comparative analysis of automated semantic clone detection techniques is performed. Semantic clones are pieces of code that fulfill the same requirement and could therefore be considered as functionally redundant. Reducing the number of semantic clones could lead to cost reduction and higher maintainability. Semantic clones were detected manually in three open source projects, and results were verified by a number of independent observers. Secondly, automated techniques were performed on the same code samples. Test characteristics like precision and recall were calculated based on comparison of automated test outcomes to manual results. Results show that in our set of test cases, matching tokens exactly is the only significant predictor for semantic clones. Including dictionary matches of words which were not directly matched did not improve performance of this test.