Ubvarol/toxic-comment-distilbert

💬 Toxic Comment Classification — DistilBERT

Detects six kinds of toxicity (toxic, severe_toxic, obscene, threat, insult, identity_hate) in online comments — a multi-label NLP problem from the Kaggle Jigsaw competition, scored by mean column-wise ROC-AUC.

A first NLP study: fine-tuning a pretrained DistilBERT (the text counterpart of image transfer learning) reaches 0.992 mean AUC. This Space is a static write-up; the training notebook lives on GitHub.