MLSNT is a multi-lingual dataset for toxicity detection created through a large language model-assisted label transfer pipeline. It enables efficient and scalable moderation across languages and platforms, and is built to support span-level and category-specific classification for toxic content.
This dataset is introduced in the following paper:
Unified Game Moderation: Soft-Prompting and LLM-Assisted Label Transfer for… See the full description on the dataset page:
https://huggingface.co/datasets/ComplexDataLab/MLSNT.
