README.md

---
license: other
language: 
- code
- en
task_categories:
- text-classification
tags:
- code
- commit_message_generation
configs:
- config_name: default
  data_files:
  - split: test
    path: data.jsonl
---

# Commit Message Quality dataset

This is the dataset for commit message quality classification, used during processing of [Commit Message Generation dataset](https://huggingface.co/datasets/JetBrains-Research/lca-commit-message-generation) from
🏟️ [Long Code Arena benchmark](https://huggingface.co/spaces/JetBrains-Research/long-code-arena).

This is a cleaned and relabeled version of the [dataset](https://zenodo.org/records/7042943#.YxG_ROzMLdo) from 📜 ["Commit Message Matters: Investigating Impact and Evolution of Commit Message Quality", ICSE'23](https://ieeexplore.ieee.org/abstract/document/10172825). We drop "Neither Why nor What" examples, clean all the external references (URLs, issues/PR references) from messages and manually label each sample with the goal of training a binary commit message quality classifier for data filtering in mind.

## How-to

Load the data via [`load_dataset`](https://huggingface.co/docs/datasets/v2.14.3/en/package_reference/loading_methods#datasets.load_dataset):

    ```
    from datasets import load_dataset

    dataset = load_dataset("saridormi/commit-message-quality", split="test")
    ```

   Note that all the data we have is considered to be in the test split.


## Dataset Structure

Each example has the following fields:

| **Field**            | **Description**                                                            |
|:---------------------|:---------------------------------------------------------------------------|
| `url`                | Link to commit on GitHub.                                                  |
| `original_message`   | Commit message as it was in the original dataset.                          |
| `message`            | Commit message cleaned from external references.                           |
| `original_label`     | Commit message label as it was in the original dataset (`Why and What`/`No Why`/`No What`). |
| `is_good`            | Whether the commit message serves as a good example of a *high quality* commit message (boolean). |
| `is_bad`             | Whether the commit message serves as a good example of a *low quality* commit message (boolean). |
| `binary_label`       | Commit message label: `1` for *high quality* messages, `0` for *low quality* messages, `null` for messages not recommended to consider for classifier training. |

Data point example:

```
{"url":"https://github.com/spring-projects/spring-boot/commit/7080500db9ecf1cf78ad23503280c713bb6e8649",
 "original_message":"Upgrade to Commons Lang3 3.6 \n  \n Closes gh-9661",
 "message":"Upgrade to Commons Lang3 3.6",
 "original_label":"Why and What",
 "is_good": False,
 "is_bad": True,
 "binary_label":0.0,
}
```