A fast and effective sentiment classification project using 3.6 million Amazon product reviews.
Built with Logistic Regression, TF-IDF, and classic preprocessing — this project demonstrates scalable NLP on real-world data.

• Source: Amazon Reviews from Kaggle (train.ft.txt)
• Size: 3,600,000 reviews
• Labels:
  • __label__1 = Negative
  • __label__2 = Positive
• Format: FastText-style plain text 📥 Download dataset from Kaggle

⚠️ Due to file size limits, amazon_train.txt and amazon_test.txt are not included in this repository.
Please download them separately and place in the project root directory if you want to re-train the model.

• Python 3.12
• Jupyter Notebook (VS Code)
• Pandas, Scikit-learn, Matplotlib, Seaborn

• Accuracy: ~90%
• F1-Score: 0.90
• Model: Logistic Regression
• Features: Top 5000 TF-IDF terms

1. Clone this repository:

   git clone https://github.com/outeast98/amazon-sentiment-classifier.git cd amazon-sentiment-classifier

2. Install dependencies:

   pip install -r requirements.txt

3. Open the Jupyter notebook:

   sentiment_analysis.ipynb

   and run each cell step by step.

This project includes a simple web interface built with Gradio, allowing users to test sentiment classification in real time.

1. Make sure you have the model and vectorizer saved as:

   • logistic_model.pkl
   • tfidf_vectorizer.pkl

2. Install Gradio:

   pip install gradio

3. Run the app:

   python app.py

The app will launch in your browser, where you can enter any product review and get an instant sentiment prediction:

😊 Positive 😠 Negative 

Yevhenii Aloshyn
Machine Learning & Cybersecurity Enthusiast
📍 Toronto, Canada
GitHub