README.md

---
dataset_info:
  features:
  - name: ecosystem
    dtype: string
  - name: vuln_id
    dtype: string
  - name: summary
    dtype: string
  - name: details
    dtype: string
  - name: aliases
    dtype: string
  - name: modified_date
    dtype: string
  - name: published_date
    dtype: string
  - name: severity
    dtype: string
  - name: score
    dtype: float64
  - name: cwe_id
    dtype: string
  - name: refs
    dtype: string
  - name: introduced
    dtype: string
  splits:
  - name: train
    num_bytes: 23841496
    num_examples: 24160
  download_size: 10220696
  dataset_size: 23841496
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

---
license: mit
---
# A Ground-Truth Dataset of Real Security Patches

This dataset provides a **ground-truth collection of real-world security patches** gathered from public GitHub repositories, cross-referenced with CVE records, and enriched with natural language artifacts and metadata. It is intended to support research in areas such as **vulnerability detection**, **security patch understanding**, **commit classification**, and **software security best practices**.


## Citation

If you use this dataset, please cite:

```bibtex
@misc{reis2021groundtruthdatasetrealsecurity,
  title={A ground-truth dataset of real security patches}, 
  author={Sofia Reis and Rui Abreu},
  year={2021},
  eprint={2110.09635},
  archivePrefix={arXiv},
  primaryClass={cs.CR},
  url={https://arxiv.org/abs/2110.09635}, 
}