CoolFace
Apppublic

DSxManash/nepali-sentiment-classifier

sourceHugging Faceupdated 9mo agoView on Hugging Face
0likes
App README

NepSentiX - Nepali Sentiment Analyzer

A machine learning-based web application for analyzing sentiment in Nepali text. The system classifies input text as Positive or Negative with a confidence score.

Live Demo

Try the application: huggingface.co/spaces/DSxManash/nepali-sentiment-classifier

Features

  • —Nepali Language Support: Designed specifically for Devanagari script
  • —Binary Classification: Positive/Negative sentiment detection
  • —Confidence Score: Probability-based confidence percentage
  • —Real-time Analysis: Instant predictions via web interface
  • —Input Validation: Handles non-Nepali text and word limits (max 100 words)

Tech Stack

ComponentTechnology
BackendFlask (Python 3.9+)
Word EmbeddingsFastText
ClassifierSupport Vector Machine (SVM)
FrontendHTML, CSS, JavaScript
DeploymentDocker on Hugging Face Spaces

Models

FastText Word Embeddings

SVM Classifier

  • —File: svm_nepali_sentiment_model.pkl
  • —Training Data: 28,800 labeled Nepali sentences (balanced dataset)
  • —Size: ~26 MB

Local Setup

Prerequisites

  • —Python 3.9 or higher
  • —~4 GB disk space for FastText model

Installation

bash
# Clone repository
git clone https://github.com/DSxManash/nepali-text-sentiment-classifier.git
cd nepali-text-sentiment-classifier

# Create virtual environment
python -m venv .venv
source .venv/bin/activate 

# Install dependencies
pip install -r requirements.txt

# Run application
python app.py

The FastText model will be automatically downloaded from Hugging Face Hub on first run.

Usage

  1. 1.Enter Nepali text in the input box (maximum 100 words)
  2. 2.Click Analyze or press Ctrl + Enter
  3. 3.View the sentiment result with confidence score

Project Structure

nepali-text-sentiment-classifier/
├── app.py                      
├── requirements.txt         
├── Dockerfile                  
├── README.md                  
├── templates/
│   └── home.html             
├── static/
│   └── style.css              
├── resources/
│   ├── nepali_sentiment_classification.ipynb  
│   └── balanced_nepali_sentiment_dataset_28800.csv
└── svm_nepali_sentiment_model.pkl

Deployment

The application is deployed on Hugging Face Spaces using Docker:

  • —FastText model is downloaded at runtime from Hugging Face Hub
  • —SVM model is bundled in the Docker image
  • —Application runs on port 7860

How It Works

  1. 1.Input Processing: Text is cleaned by removing non-Devanagari characters
  2. 2.Vectorization: FastText generates 300-dimensional word embeddings
  3. 3.Classification: SVM predicts sentiment based on averaged word vectors
  4. 4.Confidence: Sigmoid function converts SVM decision score to probability

Author

DSxManash