| --- |
| configs: |
| - config_name: default |
| data_files: |
| - split: train |
| path: data/train-* |
| license: cc0-1.0 |
| language: |
| - en |
| tags: |
| - github-issues |
| - data-science |
| --- |
| |
| # GitHub Issues Dataset |
|
|
| ## Overview |
| This dataset contains 5,000 public GitHub issues fetched using the GitHub API. It is built for data cleaning, text classification, and semantic search projects. |
|
|
| Dataset path: `duvas07/githubissues` |
|
|
| https://huggingface.co/datasets/duvas07/githubissues |
|
|
|
|
| ## Data Collection & Cleaning |
| - **Sourcing:** Fetched from public GitHub repositories using Python's `requests` package. |
| - **Cleaning:** Removed rows with missing titles (`dropna()`) and checked for duplicate IDs (`duplicated()`). |
| - **Features:** Added character length to help with text analysis. |
|
|
| --- |
|
|
| ## Privacy, Licensing & Risks |
| - **Privacy:** Data contains only public GitHub information. No private keys or passwords were included. |
| - **Licensing:** Shared under the **MIT License**. |
| - **Limitations:** Technical terms like "bug" or "crash" can be mislabeled as "negative" by standard sentiment models. Long issue text gets cut off by models with token limits. |
|
|
| --- |
| ## References |
| - **API Source:** "Data was collected from public GitHub repositories using the official GitHub REST API." |
| - **Classification Model:** `distilbert-base-uncased-finetuned-sst-2-english` |
| - **Embedding Model:** `sentence-transformers/all-MiniLM-L6-v2` |
| - **Search Library:** `faiss-cpu` |
