Ruddit is a dataset of English language Reddit comments that has fine-grained, real-valued scores for offensive language detection between -1 (maximally supportive) and 1 (maximally offensive).
The dataset was annotated using Best--Worst Scaling, a form of comparative annotation that has been shown to alleviate known biases of using rating scales.
Paper | Code | Results | Date | Stars |
---|