radastatine/bias-neutrality-corpus

Bias Neutrality Corpus

This dataset is designed for identifying and neutralizing biased language in text. It combines samples from several existing datasets along with synthetic examples to provide a comprehensive resource for bias mitigation tasks.

Dataset Structure

• Biased Sentence: The original text containing bias.
• Bias Type: Category of bias (e.g., Ageism, Gender, Religion, Racial, Ableism).
• Source Dataset: Origin of the sample (Crows Pairs, StereoSet, GBNC, Synthetic, etc.).
• Target Sentence 1-4: Neutralized versions of the biased sentence.

Splits

• train: Training data containing approximately 80% of the corpus (bias_neutrality_corpus_train.csv).
• test: Testing data for evaluation (bias_neutrality_corpus_test.csv).
• Full Corpus: The complete, unsplit dataset is available in bias_neutrality_corpus.csv.