Amazon Review Analysis using ML, NLP, and Web Scraping

Introduction

This project uses Machine Learning, Natural Language Processing (NLP), and Web Scraping in order to get real customer reviews for any product on Amazon (only amazon.com supported for now) and perform sentiment analysis that predicts whether the reviews are positive or negative.

When you run the python file amazon_review_analysis.py, you are asked to enter a product link from amazon.com (United States). After you enter the product link, the script will automatically open a Chrome window, open the product page, and surf through the product reviews. After the script goes through the desired number of review pages (by default: 5), the script outputs the total number of pages scraped, total number of reviews scraped, total number of predicted positive and negative reviews (from sentiment analysis), and average rating from the reviewers.

Setup and Deployment

1. Clone the repository into your local machine:

git clone https://github.com/nayan-pradhan/AmazonReviewsAnalysis_NLP

4. Navigate into your cloned local directory and setup a virtual environment:

pip3 install virtualenv
virtualenv venv
source venv/bin/activate

3. Install all requirement libraries:

pip3 install -r requirements.txt

4. Run the python file:

python3 amazon_review_analysis.py

Files

1. amazon_data.txt: Training data.
2. amazon_review_analysis.py: Main script that gets the url, scrapes data from the amazon website, and predicts the sentiment.
3. chromedriver: Chrome driver to open url.
4. requirements.txt: All libraries/frameworks used.
5. saved_model.pickle: Trained model that gets saved automatically.
6. script.ipynb: Jupyter Notebook script that was used in order to test functions.
7. train_model.py: Script that trains the training data and builds the training model. This scipt is automatically executed when amazon_review_analysis.py is executed.