|
|
--- |
|
|
license: mit |
|
|
task_categories: |
|
|
- text-classification |
|
|
- text-generation |
|
|
- sentence-similarity |
|
|
- text2text-generation |
|
|
language: |
|
|
- en |
|
|
tags: |
|
|
- code |
|
|
size_categories: |
|
|
- 100K<n<1M |
|
|
--- |
|
|
# GitHub Repo Metadata 5★ — Developer History and Profiling Dataset |
|
|
|
|
|
📘 Paper (FSE 2025) |
|
|
[💻 Codebase](https://github.com/SODAForge/SODAOpt) |
|
|
[📊 Source Dataset on Kaggle](https://www.kaggle.com/datasets/pelmers/github-repository-metadata-with-5-stars) |
|
|
|
|
|
## Dataset Summary |
|
|
|
|
|
This dataset provides a processed and enriched version of the "GitHub Repository Metadata with 5 Stars" dataset, reformatted to support developer modeling and task recommendation research. |
|
|
|
|
|
We provide several views of the data that are tailored for: |
|
|
- Developer-level sequence modeling |
|
|
- Socio-technical profiling |
|
|
- Text-based representation learning |
|
|
- Hybrid retrieval and recommendation tasks |
|
|
|
|
|
This dataset is used in our [FSE 2025 paper](https://doi.org/10.1145/3696630.3731437): |
|
|
**SODAOpt: Socio-Demographic and Textual Adaptive Fusion for Optimizing Developer Task Assignment**. |
|
|
|
|
|
## Dataset Structure |
|
|
|
|
|
This dataset is available in `.parquet` format and includes: |
|
|
|
|
|
| File | Description | |
|
|
|------|-------------| |
|
|
| `textual_history.parquet` | Concatenated per-user textual views of repositories | |
|
|
| `id_history.parquet` | Developer → repository interaction histories using hashed IDs | |
|
|
| `user_descriptions.parquet` | Structured per-user summaries with stars, forks, top languages | |
|
|
| `repo_info.parquet` | Clean table of repositories with hashed `repo_id` and `language_id` | |
|
|
| `mappings/item_id_map.json` | Mapping of hashed `repo_id` to sequential numeric `item_id` | |
|
|
| `mappings/language_mapping.parquet` | Mapping of language names to numeric `language_id` | |
|
|
|
|
|
## Use Cases |
|
|
|
|
|
- 🔍 Retrieval-based developer-task matching |
|
|
- 🧠 Developer embedding learning |
|
|
- 🧮 Evaluation of sequence models in software engineering |
|
|
- 🧬 Pretraining/finetuning for software-oriented LLMs |
|
|
|
|
|
## How to Cite |
|
|
|
|
|
``` |
|
|
@misc{zjkarina_2025_sodaopt, |
|
|
title = {SODAOpt: Social Dialogue Optimization Dataset}, |
|
|
author = {Karina Romanova and Sergey Senichev and Lina Veltman and Ivan Nasonov and Andrey Kuznetsov and Ilya Makarov}, |
|
|
month = {April}, |
|
|
year = {2025}, |
|
|
url = {https://huggingface.co/datasets/zjkarina/SODAOpt} |
|
|
} |
|
|
``` |
|
|
|
|
|
### Original Dataset |
|
|
> Pelmers. (2023). *GitHub Repository Metadata with 5+ Stars*. Kaggle. |
|
|
> https://www.kaggle.com/datasets/pelmers/github-repository-metadata-with-5-stars |
|
|
|
|
|
### Derived Work |
|
|
> Romanova, K., Senichev, S., Veltman, L., Nasonov, I., Kuznetsov, A., & Makarov, I. (2025). |
|
|
> *SODAOpt: Socio-Demographic and Textual Adaptive Fusion for Optimizing Developer Task Assignment*. |
|
|
> In Proceedings of the 33rd ACM International Conference on the Foundations of Software Engineering (FSE ’25). |
