Awesome

This is the replication package for "On the Evaluation of Commit Message Generation Models: An Experimental Study" accepted to ICSME 2021 and "A large-scale empirical study of commit message generation: models, datasets and evaluation" accepted to EMSE 2022.

Welcome to use our dataset, MCMD, and the evaluation scripts to test the performance of the commit message generation!

Citations for these two works can be found here.

Environment

Conda

conda create -n MCMD python=3.8 numpy=1.19.2 -y
conda activate MCMD
conda install ipykernel -y # this two lines can help jupyter notebook find the correct kernel
python -m ipykernel install --user --name MCMD --display-name "MCMD" # this two lines can help jupyter notebook find the correct kernel
pip install nltk==3.6.2 scipy==1.5.2 pandas==1.1.3 krippendorff==0.4.0 scikit-learn==0.24.1 sumeval==0.2.2 sacrebleu==1.5.1 matplotlib==3.5.1

Docker

docker pull itaowei/commit_msg_empirical

Experimental Models

CommitGen(CmtGen), NMT, CoDiSum, PtrGNCMsg, NNGen, CoRec, CodeBERT.

Experimental Datasets

• Existing Datasets: CommitGen_data, NNGen_data, CoDiSum_data

• Our Dataset: MCMD

More info about our dataset can be found here.

Evaluation Metrics

• B-Moses
• B-Norm
• B-CC
• Rouge
• Meteor

Usage demo about the metrics can be found here.

Research Questions

(RQ1) How do different BLEU variants affect the evaluation of commit message generation?

See RQ1 results here.

(RQ2) How good are the existing models and datasets?

RQ2 results: RQ2.ipynb.

(RQ3) Why do we need a new dataset MCMD for evaluating commit message generation?

RQ3 results: RQ3.ipynb.

(RQ4) What is the impact of different dataset splitting strategies?

RQ4 results: RQ4.ipynb.

(RQ5) What is the effectiveness of the pre-trained model in commit message generation?

RQ5 results: RQ5.ipynb.

Possible Ways to Improve

Evaluation results of our improvements to NNGen can be found in nngen_improvement.ipynb.

Citation

If you use this code or MCMD, please consider citing us:)

BibTex

@inproceedings{conf/icsme/TaoWSDH0ZZ21,
  author    = {Wei Tao and
               Yanlin Wang and
               Ensheng Shi and
               Lun Du and
               Shi Han and
               Hongyu Zhang and
               Dongmei Zhang and
               Wenqiang Zhang},
  title     = {On the Evaluation of Commit Message Generation Models: An Experimental
               Study},
  booktitle = {IEEE International Conference on Software Maintenance and Evolution,
               ICSME 2021, Luxembourg, September 27 - October 1, 2021},
  pages     = {126--136},
  year      = {2021},
  url       = {https://doi.org/10.1109/ICSME52107.2021.00018},
  doi       = {10.1109/ICSME52107.2021.00018}
}

@inproceedings{journals/emse/TaoWSDH0ZZ22,
  author    = {Wei Tao and
               Yanlin Wang and
               Ensheng Shi and
               Lun Du and
               Shi Han and
               Hongyu Zhang and
               Dongmei Zhang and
               Wenqiang Zhang},
  title     = {A Large-Scale Empirical Study of Commit Message Generation: Models, 
               Datasets and Evaluation},
  journal   = {Empir. Softw. Eng.},
  year      = {2022},
  doi       = {10.1007/s10664-022-10219-1}
}

RIS

Download ICSME citation, EMSE citation