AI-Driven Product Recommendation & Market Trend Analysis

📌 Project Overview

This project presents an end-to-end AI-driven Product Recommendation System built using a large-scale Amazon Products dataset (2023).

The system combines:

• Market trend analysis
• Customer behavior modeling
• Sentiment analysis (NLP)
• Emotion-aware customer segmentation
• Personalized product recommendations

The project is designed as part of Module E – AI Applications (Individual Open Project)
and demonstrates the complete AI project lifecycle, from data understanding to actionable business insights.

---

🎯 Objectives

• Perform exploratory market trend analysis on Amazon product data
• Simulate realistic customer purchase behavior
• Segment customers using RFM Analysis and Cohort Analysis
• Integrate sentiment analysis to capture emotional feedback
• Build a hybrid recommendation system with upselling and cross-selling logic
• Ensure explainability, ethics, and reproducibility

---

📊 Dataset

• Source: Amazon Products Dataset (Kaggle, 2023)
• Link: https://www.kaggle.com/datasets/lokeshparab/amazon-products-dataset

Dataset Characteristics

• ~1.1 million product records
• 140+ product category CSV files
• Includes product name, category, pricing, ratings, and review counts

⚠️ Note:
The dataset does not include customer-level purchase history or review text.
Therefore, synthetic users, transactions, and review texts were generated for academic and analytical purposes.

Due to GitHub file size limits, raw datasets are not uploaded to this repository.

---

🧠 Project Methodology

Data Understanding & Cleaning

• Combined 140+ CSV files into a single dataset
• Removed irrelevant columns
• Converted prices and ratings to numeric format
• Handled missing values using business logic

Exploratory Data Analysis (Market Trends)

• Category-wise product distribution
• Price distribution analysis
• Ratings and popularity trends
• Identification of dominant product categories

Synthetic User & Transaction Generation

• Simulated 100,000 unique users
• Generated 1,000,000 purchase transactions for the year 2023
• Created realistic purchase dates, quantities, and spending behavior

RFM Analysis (Customer Segmentation)

Customers were segmented based on:

• Recency (days since last purchase)
• Frequency (number of purchases)
• Monetary value (total spend)

Segments include:

• Big Spenders
• Loyal Customers
• At-Risk Customers
• Regular Customers

Cohort Analysis

• Grouped users by first purchase month
• Analyzed retention trends over time
• Visualized customer retention using heatmaps

Sentiment Analysis (NLP)

• Generated synthetic review text based on ratings
• Applied VADER Sentiment Analysis
• Classified sentiment as Positive, Neutral, or Negative
• Analyzed sentiment trends across product categories

Emotional Loyalty Modeling (Novel Contribution)

A new Emotional Loyalty Score was introduced by combining:

• RFM scores
• Average sentiment score per user

This enabled identification of:

• Emotionally Loyal Customers
• High-Value but Unhappy Customers
• Emotionally Disengaged Customers

Recommendation System

A hybrid recommendation engine was built:

• New users → Popular products
• Returning users → Category-based recommendations
• Big spenders → Premium upsell recommendations
• Emotion-aware recommendations using sentiment + RFM

---

📈 Key Results

• Successfully processed over 1.1 million products
• Generated large-scale synthetic customer data
• Identified meaningful customer segments and retention patterns
• Built a sentiment-aware recommendation system
• Provided explainable, business-ready insights

---

⚖️ Ethical Considerations

• No real user data was used; all customer data is synthetic
• Sentiment analysis is based on simulated review text
• Popularity and emotional bias are acknowledged and documented
• The system is intended strictly for academic and educational use

---

🚧 Limitations

• Synthetic data may not fully capture real-world intent
• Rule-based recommendation logic (no deep embeddings)
• No real-time user feedback loop
• Sentiment analysis does not capture sarcasm or complex language

---

🔮 Future Scope

• Integrate real transaction or clickstream data
• Apply collaborative filtering or deep learning models
• Use transformer-based sentiment models (BERT)
• Build interactive dashboards using Streamlit
• Explore reinforcement learning for dynamic recommendations

---

🛠️ Tech Stack

• Language: Python
• Libraries: Pandas, NumPy, Matplotlib, Seaborn
• NLP: VADER Sentiment Analysis
• Environment: VS Code, Jupyter Notebook
• Version Control: Git & GitHub

---

📁 Repository Structure

amazon-recommendation-project/

│

├── notebook/

│ └── amazon_recommendation_system.ipynb

│

├── data/

│ └── raw/ (ignored in GitHub)

│

├── README.md

└── .gitignore

---

✅ How to Run

1. Clone the repository
2. Download the dataset from Kaggle (link above)
3. Place CSV files inside data/raw/
4. Open amazon_recommendation_system.ipynb
5. Run all cells top-to-bottom

---

📌 Author

G R Shankavi Varsha

---

⭐ Final Note

This project demonstrates how AI, analytics, and NLP can be combined to build intelligent, explainable, and business-relevant systems, even when working with incomplete real-world data.