README.md

---
title: Nepali Hate Content Classification
emoji: 🔍
colorFrom: red
colorTo: blue
sdk: docker
pinned: true
license: mit
short_description: Nepali hate speech classifier with XAI
---

# Nepali Hate Content Classification System

An interactive web application for classifying hate content in Nepali social media text. Supports **Devanagari script**, **Romanized Nepali**, **English**, and **code-mixed** inputs.

## Models

| Model | HF Repo |
|-------|---------|
| XLM-RoBERTa Large | [UDHOV/xlm-roberta-large-nepali-hate-classification](https://huggingface.co/UDHOV/xlm-roberta-large-nepali-hate-classification) |
| NepaliBERT | [UDHOV/nepalibert-nepali-hate-classification](https://huggingface.co/UDHOV/nepalibert-nepali-hate-classification) |

## Classes

| Label | Description |
|-------|-------------|
| 🟢 NON_OFFENSIVE | No offensive content |
| 🟡 OTHER_OFFENSIVE | General offensive content |
| 🔴 OFFENSIVE_RACIST | Targets ethnicity, race, or caste |
| 🔴 OFFENSIVE_SEXIST | Targets gender |

## Features

- Single text and batch (CSV) classification
- Automatic script detection and preprocessing
- Emoji semantic mapping (180+ emojis)
- Confidence scores with visualization
- Explainability via LIME, SHAP, and Integrated Gradients (Captum)
- Prediction history tracking

## Project

Bachelor of Computer Engineering Final Project
Khwopa College of Engineering, Tribhuvan University, Nepal (2026)