📝 Sentiment Analysis Web App

A full-stack web application that performs sentiment analysis on restaurant reviews using machine learning. It includes a Flask backend for the API and a React frontend for user interaction.

🧰 Tech Stack

🔙 Backend

• Python: Core programming language
• Flask: Lightweight web framework for the API
• scikit-learn: Machine learning for training and inference
• nltk: Natural language processing (text cleaning & preprocessing)
• joblib / pickle: Model serialization
• matplotlib / seaborn: Visualizing evaluation metrics

💻 Frontend

• React: JavaScript library for building user interfaces
• Vite: Fast development server and bundler
• Axios: Promise-based HTTP client to interact with the API

🗃️ Data

• Dataset: Restaurant reviews (TSV format, labeled for sentiment)

🚀 Setup Instructions

🛠 Backend

cd backend
pip install -r requirements.txt         # Install dependencies
python train_model.py                   # Train and save the model (primary)
python app.py                           # Start Flask API

💻 Frontend

cd frontend
npm install                             # Install frontend packages
npx update-browserslist-db@latest      # (optional) Fix browserslist warning
npm run dev                             # Start the React app

📊 Features

• Train a sentiment classification model on restaurant reviews
• Use Logistic Regression (default) or Naive Bayes (alternative)
• Predict sentiment (positive/negative) via a REST API
• Visualize model metrics and confusion matrix
• React-based UI to input reviews and view predictions

✅ Before You Run

Make sure to:

1. Train the model using python train_model.py
2. Start the Flask backend with python app.py
3. Launch the frontend React app using npm run dev

---

🧠 Model Training Options

✔️ train_model.py (Recommended)

• Vectorizer: TF-IDF

• Model: Logistic Regression

• Usage: Best for more accurate, real-world sentiment predictions

• Saves:

  • c1_BoW_Sentiment_Model.pkl (TF-IDF vectorizer)
  • c2_Classifier_Sentiment_Model (Logistic Regression model)
  • metrics.json, analysis.png

⚙️ train_model11.py (Alternative)

• Vectorizer: CountVectorizer (Bag of Words)
• Model: Gaussian Naive Bayes
• Usage: Simpler model, quicker to train, lower accuracy
• Note: Uses only the first 900 reviews for training

💡 If you're just testing or exploring, you can try train_model11.py, but for production use train_model.py.

📌 Notes

• Ensure both backend and frontend are running on compatible ports (e.g., 5000 for Flask, 5173 for Vite).
• CORS configuration may be required if cross-origin requests are blocked.
• Model files (*.pkl, classifier) must be generated before API calls will work.
• After training, metrics are saved in metrics.json and confusion matrix in analysis.png.