Boosting Patch Correctness Checking via Combining Code-Edit Embeddings as Additional What-to-Change Information

Chontipan Plengvittaya, Zee Hen Tang, Mi-Yen Yeh, Chih-Ya Shen

Abstract: Patch Correctness Checking is the task of detecting automated-program-repair-generated patches that can pass the test suite but are still buggy. In this study, we investigated a new feature to improve the classification performance for automated patch correctness checking. Our approach involved leveraging code-change information at a finer granular level by embedding the code edits and combining them with existing code-change vectors. Through extensive experiments, the results revealed gains up to 3.4% and 3.2% on the Fl-score and AUC.