MilaNLProc/njh-classifier

Not Just Hate (NJH)

Uploaded Version

The uploaded balanced model for multi-label harmful speech classification; see the paper for more details.

Labels: Profanity, Insults, Outrage, Character Assassination, Discrimination, Hostility, Incivility, and Intolerance.

Label Details

Incivility refers to discourse that is rude, offensive, or use of swear-words/cursing. This includes four sub-types:

• Profanity: The use of swear words and vulgar language, not necessarily targeted at individuals/groups
• Insults: Direct offensive language (e.g. belittling, demeaning, vulgar, and/or insulting language) targeted at individuals or groups, including their character, behaviors, ideas, or opinions.
• Outrage: Expressions of extreme anger or indignation, using dramatic/emotional/provocative language or negative exaggeration when referencing behaviors, policies, views, political opinions, etc. of a person, group, branch of government, party or organization.
• Character Assassination: Attacks on someone's moral character, credibility, honesty, or integrity. Character assassination can target individuals or organizations (e.g. NGOs, Government), but not identity-based groups (coded as intolerance/discrimination)

Intolerance refers to expressions that threaten democratic norms and values, such as equality, diversity, and freedom. This includes two sub-types:

• Discrimination: Content that discriminates against individuals or groups based on their identity
• Hostility: An aggregated label that encompasses several particularly harmful types of speech including "hateful speech, dehumanization, serious threat-personal abuse-harassment, and democratic threat"

Cookbook

If you open the Files section of this repo you will find the annotation cookbook.

Cite As

Bianchi, F., Hills, S., Rossini, P., Hovy, D., Tromble, R., & Tintarev, N. (2022). "It's not just hate": a multi-dimensional perspective on detecting harmful speech online. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics.