---
title: Toxic Comment Classifier
emoji: 🛡️
colorFrom: gray
colorTo: red
sdk: gradio
sdk_version: 6.14.0
app_file: app.py
pinned: true
license: mit
---

# 🛡️ Toxic Comment Classifier

**NLP Final Project** — Fine-tuned DistilBERT on the Jigsaw Toxic Comment Classification Dataset.

## What it does
Classifies English text comments into **6 toxicity categories**:
- ☠️ Toxic
- 💀 Severe Toxic
- 🤬 Obscene
- ⚔️ Threat
- 😤 Insult
- 🎯 Identity Hate

## Model
- **Base**: `distilbert-base-uncased`
- **Training data**: ~160K comments from the [Jigsaw dataset](https://www.kaggle.com/c/jigsaw-toxic-comment-classification-challenge)
- **Fine-tuning**: Multi-label classification with BCEWithLogitsLoss
- **Target F1**: > 85% micro-average

## How to use
1. Type or paste any English comment in the text box
2. Click **Analyze**
3. See scores per toxicity category + overall verdict

## Tech Stack
`transformers` · `PyTorch` · `Gradio` · `HuggingFace Hub`