Kevin: Knowledge-Enhanced Conventional Commit Message Generation
收藏NIAID Data Ecosystem2026-05-02 收录
下载链接：
https://zenodo.org/record/11523050
下载链接
链接失效反馈官方服务：
资源简介：
Introduction
We purpose a novel conventional commit message generation framework Kevin. In this directory, we give all code, dataset, and experiment results.
We use the dataset MCMD to evaluate our model, and filter some meaningless data on [`it`](dataset/MCMD_filter.zip).
Now we introduce the purpose of each folder:
1. `code`: this is the implement of our approach.
1. `commit_type_classification`: this is implement of our approach in commit type classification, and there are detailed documents in this folder.
2. `commit_description_classification`: this is implement of our approach in commit description generation, and there are detailed documents in this folder.
2. `dataset`: this is the dataset we used to evaluate our approach, `MCMD_filter.zip` contains the filtered version of the dataset MCDM, `metrics.zip` includes 10 code change metrics used in commit type classification.
3. `empirical_study`: this is the result of empirical study abour the correlation between code change situations and 8 commit types.
4. `experiment_results`: this is all original experiment results of Kevin and six state-of-the-art approaches.
1. `RQ1 Overall Effectiveness`: this is for experiment RQ1.
2. `RQ2 Ablation Study`: this is for experiment RQ2.
3. `RQ3 Human Evaluation`: this is for experiment RQ3.
4. `metrics`: this is the implement of automated evaluation metrics BLEU, ROUGE-L, and METEOR.
5. `checkpoints`: this directory includes five checkpoints of commit description generation model across five languages.
# Usage
Before run the model, unzip all dataset in directory `dataset/`.
``` shell
unzip MCMD_filter.zip
unzip metrics.zip
```
In the directory `code/commit_type_classification` and `code/commit_description_generation`, there are detailed document for install dependencies and run the model.
To run commit type classification model,
``` shell
python classification.py
```
If extract features from original dataset, configure parameters in function `main()` of `code/commit_type_classification/get_feature.py`, and
``` shell
python get_feature.py
python classification.py
```
To run commit description generation model, prepare the data modify the parameters in function `main()`.
``` shell
python prepare.py
```
configure parmeters in directory `code/commit_description_generation/conf/*.py`, includes the path of dataset, model, and hyperparameters.
``` shell
CUDA_VISIBLE_DEVICES=0 python eval.py +model=codet5
```
If train the model,
``` shell
CUDA_VISIBLE_DEVICES=0 python train.py +model=codet5
```
创建时间：
2024-09-12
