README.md

metadata

license: mit
tags:
  - ai
  - modeltraining
  - machinelearning
language:
  - en
pretty_name: 'AI code review benchmark dataset '

PR_Review-Benchmark-Dataset

A High-Signal Dataset for Evaluating AI Code Review & Test Generation Systems

Overview

PR_Review-Benchmark-Dataset is a curated dataset of merged GitHub Pull Requests, code patches, and human review comments, designed for benchmarking:

1. AI Code Review Systems
2. Test Case Generation Models
3. Software Engineering LLMs

All data is derived from publicly available open-source repositories under permissive licenses and processed using a privacy-preserving pipeline. This dataset is suitable for academic research, open benchmarking, and commercial evaluation.

Dataset Contents

Each entry contains:

1. Pull request metadata
2. Code patches
3. Human review comments
4. Anonymized contributor identifiers
5. Repository license information
6. Derived evaluation metadata

Example:

```json
{
  "_id": "...",
  "repo": "psf/requests",
  "license_spdx": "Apache-2.0",
  "title": "...",
  "files": [...],
  "reviews": [...],
  "review_comments": [...],
  "meta": {...}
}

Schema Overview

Field	Description
_id	Unique salted hash identifier
_schema_version	Dataset schema version
_cleaned_at	Processing timestamp
repo	Repository name
license_spdx	SPDX license identifier
files	Modified files with patches
reviews	Pull request reviews
review_comments	Inline reviewer comments
meta.touches_tests	Test-modifying PR flag
meta.meaningful_comment_count	Review signal strength
meta.review_state_summary	Review status summary

Legal & Privacy Compliance

This dataset follows a Privacy by Design approach and complies with applicable data protection regulations. 🇮🇳 India: Digital Personal Data Protection Act (DPDP), 2023 Data is processed under Section 3(c)(ii) exemption: Publicly available personal data made available by the data principal.

Privacy Measures

Measure	Description
Anonymization	Usernames are replaced with salted SHA-256 hashes
Redaction	Names, emails, URLs, and PII are removed
Content Filtering	Private/non-public repos excluded
License Filtering	Only permissive licenses allowed

Data Processing Pipeline

1. Multi-pass regex filtering
2. NLP-based name detection
3. Stable pseudonym mapping
4. Patch-level sanitization
5. Deduplication No original personal identifiers are retained.

Attribution & Provenance

Dataset Attribution Manifest

Generated on: 2026-02-12 This dataset contains code snippets, pull request metadata, and human review comments derived from the following public open-source repositories. All data was collected from repositories explicitly using permissive licenses.

Repositories Included

#	Repository	License	URL
1	psf/requests	Apache-2.0	https://github.com/psf/requests
2	django/django	BSD-3-Clause	https://github.com/django/django
3	tiangolo/fastapi	MIT	https://github.com/tiangolo/fastapi
4	numpy/numpy	BSD-3-Clause	https://github.com/numpy/numpy
5	pandas-dev/pandas	BSD-3-Clause	https://github.com/pandas-dev/pandas
6	scikit-learn/scikit-learn	BSD-3-Clause	https://github.com/scikit-learn/scikit-learn
7	pytorch/pytorch	BSD-3-Clause	https://github.com/pytorch/pytorch
8	tensorflow/tensorflow	Apache-2.0	https://github.com/tensorflow/tensorflow
9	python/cpython	PSF License	https://github.com/python/cpython

Legal Notice

All original code is copyright (c) its respective contributors. This dataset is a derived work for machine learning research and benchmarking. No ownership of original content is claimed.

Limitations & Biases

1. Focuses on large, mature OSS projects
2. Underrepresents small/private repos
3. English-centric discussions
4. Limited to merged PRs

This dataset reflects real-world OSS workflows, not all software development contexts.

Recommended Citation

• PR Review Benchmark Dataset (2026).
• Curated for Code-LLM Evaluation.

Responsible Use

This dataset is intended for:

• Improving developer tools
• Advancing ML research
• Supporting open-source ecosystems

Users must not:

• Attempt deanonymization
• Extract personal identities
• Misrepresent authorship
• Violate upstream licenses

Contact & Maintenance

For issues, improvements, or corrections:

• Open a GitHub Issue
• Submit a Pull Request