Movie Sentiment Analyzer

A fullstack web application that classifies movie reviews as Positive or Negative using a fine-tuned DistilBERT model with LoRA.

Project Overview

This project demonstrates an end-to-end machine learning pipeline:

1. Fine-tuned DistilBERT on the IMDB dataset using LoRA (Low-Rank Adaptation)
2. Built a Flask API to serve predictions
3. Created a SvelteKit frontend for user interaction

Tech Stack

Component	Technology
Model	DistilBERT + LoRA
Training	Google Colab, Hugging Face Transformers, PEFT
Backend	Flask, PyTorch
Frontend	SvelteKit, Tailwind CSS

Project Structure

sentiment-app/
├── server.py                 # Flask API
├── my-sentiment-model/       # LoRA adapter files
│   ├── adapter_config.json
│   ├── adapter_model.safetensors
│   └── tokenizer files
└── frontend/                 # SvelteKit app
    └── src/
        └── routes/
            └── +page.svelte

Model Details

• Base Model: distilbert-base-uncased
• Fine-tuning Method: LoRA (Parameter-Efficient Fine-Tuning)
• Dataset: IMDB Movie Reviews (25,000 training samples)
• Task: Binary Sentiment Classification (Positive/Negative)
• Trainable Parameters: 739,586 (1.09% of total)
• Training Time: ~4 minutes on Google Colab

Training Results

Epoch	Training Loss	Validation Loss
1	0.3236	0.3291
2	0.2881	0.3239
3	0.2659	0.3179

Setup Instructions

1. Clone the repository

git clone https://github.com/YOUR_USERNAME/sentiment-app.git
cd sentiment-app

2. Download the model

Download the LoRA adapter files and place them in the my-sentiment-model/ folder.

3. Install dependencies

pip install flask flask-cors transformers torch peft

4. Run the backend

python server.py

The API will be available at http://127.0.0.1:5000

5. Run the frontend

cd frontend
npm install
npm run dev

The app will be available at http://localhost:5173

API Reference

Analyze Sentiment

Endpoint: POST /api/analyze

Request:

{
  "text": "This movie was absolutely fantastic!"
}

Response:

{
  "sentiment": "Positive",
  "confidence": 97.85
}

Health Check

Endpoint: GET /api/health

Response:

{
  "status": "ok"
}

How It Works

User Input --> SvelteKit Frontend --> Flask API --> LoRA Model --> Prediction
     |                                                                |
     <-------------------------- Response ---------------------------

1. User enters a movie review in the frontend
2. Frontend sends POST request to Flask API
3. API tokenizes text and runs it through the model
4. Model returns sentiment prediction with confidence score
5. Frontend displays the result

Limitations

• Trained only on movie reviews; may not generalize to other domains
• Binary classification only (no neutral option)
• Maximum input length of 512 tokens
• English language only

Future Improvements

• Deploy to Hugging Face Spaces
• Add multi-class sentiment (positive, neutral, negative)
• Support for other languages
• Batch analysis for multiple reviews

License

MIT

Author

Michael Fafore

Acknowledgments

• Hugging Face for Transformers and PEFT libraries
• IMDB dataset
• Google Colab for free GPU access