Sentiment Analysis of Product Reviews

This repository contains the implementation of a sentiment analysis system designed to classify e-commerce product reviews as positive or negative. The project combines machine learning, transfer learning, and web development techniques to provide an end-to-end solution for real-time sentiment analysis.

---

Overview

Understanding customer sentiment in product reviews is crucial for businesses and consumers. This project aims to address this challenge by:

1. Developing an initial model using a Recurrent Neural Network (RNN) for sentiment classification.
2. Improving accuracy using transfer learning with a pre-trained BERT model.
3. Deploying the final model in a Flask-based web application that provides real-time predictions for user-input Amazon product URLs.

---

Technologies Used

Programming Language

• Python

Machine Learning Libraries

• TensorFlow
• PyTorch
• Hugging Face Transformers

Web Development

• Flask

Web Scraping

• BeautifulSoup
• Scrapy

Visualization

• Matplotlib
• Seaborn

---

Project Features

1. Data Collection

   • Datasets of reviews were collected from Amazon, Yelp, and IMDb.
   • Additional reviews were scraped using Python tools.

2. Preprocessing

   • Text data was cleaned using tokenization, stop word removal, and lemmatization.

3. Model Development

   • A custom RNN was initially trained but achieved low accuracy (~60%).
   • A pre-trained BERT model was fine-tuned, achieving 92% accuracy.

4. Web App Deployment

   • A Flask-based app was developed where users can input an Amazon product URL.
   • The app scrapes reviews from the product page and performs real-time sentiment analysis.

---

Model Performance

Metric	RNN	BERT
Accuracy	~60%	92.18%
Precision	-	89.45%
Recall	-	95.64%
F1-Score	-	92.45%

---

How to Use

Prerequisites

• Python 3.8 or later
• Install required libraries using the following:

  pip install -r requirements.txt

Run the Web App

1. Clone the repository:

   git clone https://github.com/username/sentiment-analysis
   cd sentiment-analysis

2. Run the Flask app:

   python app.py

3. Open your browser and navigate to http://127.0.0.1:5000.

4. Input an Amazon product URL to get a sentiment analysis report.

Future Work

• Mitigate overfitting in the BERT model using regularization techniques.
• Expand the dataset to include reviews from more diverse domains.
• Explore more advanced transfer learning models like GPT or RoBERTa.

---

Authors

• Kevin Igweh
• Bruna Jacinto Grassi

---

Acknowledgments

We thank Dr. Ajmery Sultana for her guidance and support throughout this project.

---