An Empirical Study on Commit Message Generation using LLMs via In-Context Learning

This repository contains the necessary data and code to reproduce the work done in our paper.

Dataset

• MCMD: https://zenodo.org/record/7196966#.Y0juJHZBxmM
• MCMD-New: https://drive.google.com/file/d/1zxjxX1nYcGOT2z2OWxDx_doiLS07MhyJ/view?usp=drive_link

Baselines

We reused the code and the hyperparameter values released by the authors in the GitHub repositories for all baselines.

• NNGen: https://github.com/Tbabm/nngen
• CCT5: https://github.com/Ringbo/CCT5
• COME: https://github.com/hyc2026/come

Experiment

• RQ1.py: the script for RQ1
• RQ2.py: the script for RQ2

Results

Detailed results on DeepSeek-V2 for RQ1 can be found in figures/ directory.