🧠 NLP-Powered Review Analysis Project

This project leverages Natural Language Processing (NLP) and Transformer-based models to analyze Amazon product reviews. It covers three main tasks: sentiment classification, product category clustering, and review summarization using generative AI.

---

📌 1. Review Classification

🎯 Objective

Classify customer reviews into positive, negative, or neutral categories to help the company improve its products and services.

🛠️ Task

Develop a model that categorizes textual reviews into sentiment classes using pre-trained transformer-based models.

✅ Approach

• Used Transformer-based models (e.g., BERT) via Hugging Face to leverage powerful language representations.
• Focused on fine-tuning rather than training from scratch.

📊 Results

• Test Accuracy: 0.967
• Test Precision: 0.966
• Test Recall: 0.967
• Test F1-Score: 0.967

Classification Report: precision recall f1-score support Negative 0.86 0.84 0.85 337 Neutral 0.66 0.64 0.65 261 Positive 0.99 0.99 0.99 6069

Accuracy 0.97 6667

---

📌 2. Product Category Clustering

🎯 Objective

Simplify the product dataset by clustering detailed product types into 4–6 broader meta-categories.

🛠️ Task

Build a model that analyzes product names/descriptions and assigns them to high-level categories.

✅ Categories Identified

After analyzing the dataset, the following meta-categories were created:

Meta Category	Number of Reviews
Batteries & Accessories	12,071
Tablets & E-readers	6,486
Fire Tablets & Kindle Devices	6,307
Kids' Tablets & Educational Devices	6,084
Smart Devices & Alexa	2,384

---

📌 3. Review Summarization Using Generative AI

🎯 Objective

Generate blog-style summaries per product category that highlight:

• Top products and their key differences
• Top complaints for each product
• Worst product and why it should be avoided

🛠️ Task

Develop a pipeline that automatically generates short articles from customer review data.

✅ Approach

✅ Load Pre-trained Models

• facebook/bart-large-cnn for summarization
• facebook/bart-large-mnli for zero-shot classification

🧩 Summarize Complaints

• summarize_reviews() condensed top negative reviews per product.

🏷️ Extract Key Features

• extract_features() used zero-shot classification with labels like:
  • Battery
  • Performance
  • Durability
  • Price

🌟 Identify Top & Worst Products

• Selected Top 3 products based on positive reviews.
• Identified Worst product using negative review count.

📝 Generate Articles

For each meta-category:

• Introduced top 3 products with their advantages and complaint summaries.
• Included the worst product and why it should be avoided.

📄 Output

The generated summaries act as automated product recommendation articles to help customers make informed purchasing decisions.

---

🚀 Tools & Libraries Used

• Python
• Hugging Face Transformers
• Scikit-learn
• Pandas / NumPy
• Jupyter Notebook