moderation-abuse-korean

Purpose

This model detects hate speech in Korean text. Outputs four concepts:

• "3" : Hate
• "2" : Offensive
• "1" : Gender bias
• "0" : Other bias

Architecture

This is a KcELECTRA model that is finetuned for hate speech detection.

The original KcELECTRA-base model is a pretrained ELECTRA model that is trained from scratch using comments from Naver News.

The dataset used for finetuning:

• BEEP! Korean Corpus of Online News Comments for Toxic Speech Detection

Intended Use

Multi-class classification of Korean text

Resources

• This model on HuggingFace
• Original ELECTRA paper:
  • ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators
• ELECTRA on GitHub
• KoELECTRA on GitHub
• KcELECTRA finetuning on GitHub