🤖 Sentiment Analysis AI App

A complete end-to-end sentiment analysis system powered by a fine-tuned Transformer model. This app analyzes English text and classifies it as positive, negative, or neutral using a real-time web interface. It includes:

• HuggingFace Transformer fine-tuning
• FastAPI backend for inference
• Streamlit frontend with animated UI
• Dockerized deployment

Features

• Fine-tuned cardiffnlp/twitter-roberta-base-sentiment-latest on 3-label data

• Real-time predictions via FastAPI (/predict)

• Streamlit UI with:

  • Emoji-based label
  • Confidence bar animation
  • Flip-in effect on sentiment change

• Fully containerized with Docker Compose

---

Project Structure

Sentiment_Analysis_AI/
├── backend/
│   ├── app.py                # FastAPI app
│   ├── model.py              # Model loading logic
│   ├── model/                # Fine-tuned model saved here
│   └── __init__.py
├── ui.py                    # Streamlit frontend
├── finetune.py              # Fine-tuning script
├── run_all.py               # Local launcher script
├── Dockerfile.backend
├── Dockerfile.frontend
├── docker-compose.yml
├── requirements.txt
└── README.md

---

Setup Instructions

Requirements

• Python 3.10+
• Docker & Docker Compose (if containerized)

---

Local Development

1. Install Dependencies

pip install -r requirements.txt

2. Run Backend

uvicorn backend.app:app --reload --port 8000

3. Run Frontend (separate terminal)

streamlit run ui.py

Then go to http://localhost:8501.

---

Dockerized Deployment

1. Build and Run All

docker-compose up --build

2. Open in Browser

• UI: http://localhost:8501
• API Docs: http://localhost:8000/docs

---

Fine-tuning

Train a custom sentiment classifier:

python finetune.py \
  --data data/files/train_converted.jsonl \
  --epochs 3 \
  --lr 3e-5

Outputs saved in backend/model/

---

Example API Call

curl -X POST http://localhost:8000/predict \
     -H "Content-Type: application/json" \
     -d '{"text": "This product is awesome!"}'

Response:

{
  "label": "positive",
  "score": 0.9823
}

---

Troubleshooting

Issue	Fix
ConnectionError in UI	Use http://backend:8000 inside Docker, not localhost
Docker timeout	Use --default-timeout=300 in Dockerfile
CORS/XSRF warning	Safe to ignore unless deploying cross-domain

---

Author

Sanjay Ravichander Junior Machine Learning & Deep Learning Engineer

---

License

MIT

---