t-davidson/hate-speech-and-offensive-language

Includes a hand-labeled dataset of ~16K tweets classified into hate speech, offensive language, and neither, plus a custom lexicon and Python classifier for detecting hate speech on new data. Addresses the critical distinction between hate speech and general offensive language through multi-class annotation rather than binary classification. The repository also documents racial bias issues in hate speech detection datasets, highlighted in a companion 2019 paper.

837 stars. No commits in the last 6 months.