PurposeA small, focused binary-labeled dataset to support the identification, measurement, and mitigation of racial, cultural, and related social biases in language models and NLP systems.
Each example includes:
A potentially controversial text prompt or statement (question/claim about race, culture, attraction, immigration, heritage, hiring, etc.)
A binary label:
1 = generally acceptable / "it's okay"
0 = generally problematic / "it's not… See the full description on the dataset page:
https://huggingface.co/datasets/AjourDev/racial_bias.
