metadata

license: mit
task_categories:
  - text-classification
language:
  - en
size_categories:
  - 1K<n<10K

🧠 OpenAI Moderation Binary Dataset

This dataset is a binary-labeled version of the original OpenAI Moderation Evaluation Dataset, created to support safe/unsafe classification tasks in content moderation, safety research, and AI alignment.

📦 Dataset Details

Original Source: OpenAI Moderation API Evaluation Dataset
License: MIT (inherited from original repo)
Samples: 1,680 total
Labels:
- "safe" (no harm labels present)
- "unsafe" (at least one moderation label present)

📁 Structure

Each row consists of:

{
  "prompt": "Some user input text...",
  "prompt_label": "safe"  // or "unsafe"
}

🧹 Preprocessing

This version was derived by:

Downloading and parsing the original JSONL dataset (samples-1680.jsonl.gz)
Creating a new column called prompt_label, based on the presence of any of the following 8 moderation labels:
- S (sexual)
- S3 (severe sexual)
- H (hate)
- H2 (severe hate)
- V (violence)
- V2 (severe violence)
- HR (harassment)
- SH (self-harm)
Assigning:
- prompt_label = "unsafe" if any of those were 1
- prompt_label = "safe" if all were 0
Removing the original moderation columns, leaving only:
- prompt
- prompt_label

📊 Label Distribution

Label	Count	%
`safe`	1158	~68.9%
`unsafe`	522	~31.1%

💡 Intended Use

This dataset is designed for:

Binary classification (safe vs unsafe prompt detection)
Content moderation and safety evaluation
Educational and research purposes

📚 Citation

If you use this dataset, please cite the original authors of the OpenAI Moderation dataset:

OpenAI (2022).
A Holistic Approach to Undesired Content Detection in the Real World.
https://github.com/openai/moderation-api-release

🙏 Acknowledgements

Huge credit to OpenAI for releasing the original dataset.

This binary-labeled version was created for ease of evaluation and validation.