Product Review Sentiment Analysis

An end-to-end NLP project that classifies product reviews as Positive or Negative using TF-IDF features and a Linear SVM model, deployed as an interactive Streamlit web application.

---

📌 Overview

This project demonstrates a complete sentiment analysis workflow:

• Data ingestion from the Amazon Fine Food Reviews dataset (Kaggle).
• Text preprocessing using NLTK (tokenization, stopword removal, lemmatization).
• Feature extraction using TF-IDF (unigrams + bigrams).
• Model training with Linear SVM (and Logistic Regression benchmark).
• Model persistence (sentiment_model.pkl, vectorizer.pkl) with joblib.
• Web UI for single review and bulk CSV predictions using Streamlit.
• Ready for deployment on Streamlit Community Cloud.

---

🧠 Architecture

1. Data Layer

   • Raw reviews from Kaggle (data/Reviews.csv).
   • Columns Text and Score used for sentiment modeling.

2. Preprocessing & Feature Layer

   • Custom text cleaner:
     • lowercase
     • HTML & URL removal
     • digits & punctuation removal
     • stopword removal
     • lemmatization
   • TF-IDF vectorization (max_features=50,000, ngram_range=(1,2)).

3. Model Layer

   • Binary labels:
     • Score 4–5 → positive
     • Score 1–2 → negative
   • Neutral (Score = 3) reviews removed.
   • Models evaluated:
     • Logistic Regression
     • Linear SVM (chosen as final model).

4. Deployment Layer

   • app.py (Streamlit app) for:
     • Single review prediction
     • CSV upload (review column) for bulk prediction
   • Hosted locally or on Streamlit Community Cloud.

---

📂 Project Structure

product-review-sentiment-analysis/
├── app.py
├── requirements.txt
├── README.md
├── data/
│   └── Reviews.csv
├── models/
│   ├── sentiment_model.pkl
│   └── vectorizer.pkl
└── notebooks/
    └── product_review_sentiment.ipynb