159,571 Wikipedia talk page comments labeled across six distinct categories of toxicity including toxic, severe_toxic, and identity_hate. Each record contains raw text from human discussions paired with binary indicators of offensive behavior as determined by human raters.
Use Cases
- Train a multi-label text classifier using the comment_text and the six toxicity category columns
- Analyze linguistic markers of aggression using the identity_hate and threat labels
- Develop content moderation algorithms to filter comments based on the obscene and insult scores
Strengths
- 159,571 rows of Wikipedia comment text
- Six binary label columns: toxic, severe_toxic, obscene, threat, insult, and identity_hate
- Human-labeled ground truth for subjective text classification