Ghost Annotator framework for label variation in content moderation

Arxiv CS.CL (NLP/language models)

Framework uses conformal prediction to model human label variation in content moderation. It addresses uncertainty in LLM-annotated data and compares agreement metrics.

conformal-predictioncontent-moderation

Arxiv CS.CL (NLP/language models)·Mirko LaiAlessandra UrbinatiSimona FrendaFabiana VerneroMarco Antonio Stranisci

8 days ago