| Presentation | 2024-03-07 For evaluating the effectiveness of CodeT5 transfer learning in refactoring recommendations. Yuto Nakajima, Kenji Fujiwara, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | Refactoring is "the process of restructuring the internal architecture of software to make it easier to understand and modify without changing its external behavior," and it is a crucial activity in software development. In this study, we utilized CodeT5, a pre-trained model specialized in source code, to perform transfer learning for refactoring recommendations, aiming to improve prediction accuracy when using deep learning models. Specifically, we fine-tuned CodeT5 as a recommendation model using a dataset comprised of over 300,000 Extract Method refactoring from 11,149 real projects across three repository groups: Apache, F-Droid, and GitHub. The model's accuracy was evaluated using Precision, Recall, and F-measure as metrics. The results showed that the model could identify 97% of the methods targeted for refactoring from the recommendation model, achieving a 13 percentage point increase in precision and a 12 percentage point increase in recall compared to the method by Aniche et al. This indicates that transfer learning using CodeT5 is effective for refactoring recommendations. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Refactoring recommendation / deep learning / CodeT5 / pre-training model |
| Paper # | SS2023-62 |
| Date of Issue | 2024-02-29 (SS) |
| Conference Information | |
| Committee | SS |
|---|---|
| Conference Date | 2024/3/7(3days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | Kozo Okano(Shinshu Univ.) |
| Vice Chair | Yoshiki Higo(Osaka Univ.) |
| Secretary | Yoshiki Higo(Shinshu Univ.) |
| Assistant | Shinsuke Matsumoto(Osaka Univ.) |
| Paper Information | |
| Registration To | Technical Committee on Software Science |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | For evaluating the effectiveness of CodeT5 transfer learning in refactoring recommendations. |
| Sub Title (in English) | |
| Keyword(1) | Refactoring recommendation |
| Keyword(2) | deep learning |
| Keyword(3) | CodeT5 |
| Keyword(4) | pre-training model |
| 1st Author's Name | Yuto Nakajima |
| 1st Author's Affiliation | Tokyo City University(Tokyo City University) |
| 2nd Author's Name | Kenji Fujiwara |
| 2nd Author's Affiliation | Tokyo City University(Tokyo City University) |
| Date | 2024-03-07 |
| Paper # | SS2023-62 |
| Volume (vol) | vol.123 |
| Number (no) | SS-414 |
| Page | pp.pp.79-84(SS), |
| #Pages | 6 |
| Date of Issue | 2024-02-29 (SS) |