moderation-abuse-japanese

Purpose

This model detects cyberbullying in Japanese. Outputs two concepts:

• LABEL0 (cyberbullying not detected)
• LABEL1 (cyberbullying detected)

Pro Tip:

You can perform audio moderation by creating a workflow that converts audio to text (such as the Facebook Wav2Vec2 model finetuned on Japanese), and then use this Japanese text moderation model on that generated text.

Architecture

This is an ELECTRA Small model finetuned for cyberbullying detection. The original model was pretrained using ELECTRA Small model settings. https://arxiv.org/abs/2003.10555

The two datasets used for finetuning:

• Harmful BBS Japanese comments dataset
• Twitter Japanese cyberbullying dataset

Intended Use

Classification of Japanese text

Resources

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