🛒 E-Commerce Product Recommendation System (Enhanced Version)

A Flask-based product recommendation system that analyzes user purchase history from CSV files and recommends relevant products based on category preferences, brand affinity, discounts, popularity, and similarity scoring.

This project is designed to be simple, explainable, and demo-ready, suitable for:

• Mini-project / AI-ML integration demo
• Backend recommendation logic showcase

---

📌 Problem Statement

E-commerce platforms need to recommend products that match a user's interests while also considering brand loyalty, product popularity, and discounts.

In this project, we:

1. Analyze past user orders stored in a CSV file
2. Identify the user’s interest patterns (categories, brands, frequency, recency)
3. Recommend new products from another CSV file
4. Compute a hybrid similarity score (category + brand + popularity + discount)
5. Display results in a web dashboard using Flask + Jinja2

All data is handled offline using CSV files (no database required).

---

🧠 How the System Works (High Level)

User selects Customer ID
        ↓
System reads users_order.csv
        ↓
Finds dominant categories, brands & purchase behavior
        ↓
Matches with all_products.csv
        ↓
Computes hybrid similarity score
        ↓
Displays recommendations with explainability

---

📂 Project Structure

project-root/
│
|__ product_images
├── app.py                    # Flask application
├── recommender.py            # Recommendation logic (hybrid scoring)
│
├── data/
│   ├── users_order.csv       # User purchase history
│   └── all_products.csv      # Product catalog with brand, rating, reviews
│
├── templates/
│   ├── index.html            # User selection page
│   ├── recommendations.html  # Recommendation dashboard
│   ├── products.html         # Full catalog view
│   └── users.html            # All users & purchase histories
│
├── static/
│   └── images/               # Optional local images
│
└── README.md

---

📊 Dataset Description

1️⃣ users_order.csv (User Purchase History)

Each row represents one product purchase by a user.

Columns:

• customer_id – Unique user ID
• product_id – Product identifier
• product_name – Name of product
• product_category – Category (Electronics, Fashion, etc.)
• quantity – Quantity purchased
• price_usd – Price per unit
• payment_method – Credit Card / Debit Card / Paypal / UPI
• order_date – Purchase date
• total_amount – Final order value

✔ Users can purchase multiple categories and brands over time

---

2️⃣ all_products.csv (Product Catalog)

Contains all available products on the platform.

Columns:

• product_id
• product_name
• product_category
• price
• discount_price
• stock_quantity
• image_url (public real image URLs)
• brand
• rating (1–5 stars)
• reviews_count (number of reviews)

---

🤖 Recommendation Logic (Core Idea)

The system uses a hybrid content-based filtering approach:

🔹 Step 1: User Profiling

• Count category-wise purchases
• Identify dominant brands & categories
• Consider quantity and recency

🔹 Step 2: Product Matching

• Recommend products from:
  • Frequently bought categories
  • Same brands or related brands
• Exclude already purchased products

🔹 Step 3: Hybrid Similarity Score

Each recommended product gets a score based on:

• Category match (cosine similarity)
• Brand affinity
• Popularity (rating + reviews)
• Discount attractiveness
• Recency bonus

Example:

Final Score = 0.5 * Category Similarity
            + 0.2 * Popularity
            + 0.2 * Discount
            + 0.1 * Diversity Factor

This score is shown in the UI for explainability.

---

🖥️ Web Application Flow

1️⃣ User selects Customer ID from dropdown

2️⃣ System shows:

• Previously purchased products
• Recommended products

3️⃣ Recommendation cards include:

• Product image
• Name, category, brand
• Price (with discount)
• Rating & reviews
• Similarity score badge

---

⚙️ Technologies Used

• Python 3
• Flask – Backend web framework
• Pandas – Data processing
• Scikit-learn – Cosine similarity
• Jinja2 – HTML templating
• Bootstrap 5 – Frontend styling
• CSV Files – Data storage

---

▶️ How to Run the Project

pip install flask pandas scikit-learn
python app.py

Then open:

http://127.0.0.1:5000/

---

🎓 Academic Value (For HOD / Faculty)

✔ Demonstrates real-world recommendation logic
✔ Clear separation of data, logic, and UI
✔ Explainable AI (hybrid similarity scores)
✔ No heavy infrastructure required
✔ Easily extensible to ML models

---

🚀 Future Enhancements

• Collaborative filtering
• ML-based embeddings (Word2Vec, BERT)
• User clustering
• Database integration (MySQL / MongoDB)
• Login-based personalization
• Chart.js analytics (category distribution, score breakdown)

---

📌 Conclusion

This project simulates a real e-commerce recommendation engine using hybrid scoring (category + brand + popularity + discount), making it ideal for learning, demos, and academic evaluation.