🛍️ Marketplace Recommendation Engine

Hybrid AI Recommendation System for Smart Product Discovery

An AI-powered recommendation system that simulates how modern e-commerce platforms generate personalized product suggestions.

This project combines:

• Collaborative Filtering (SVD Matrix Factorization)
• Content-Based Filtering (TF-IDF + Cosine Similarity)

into a Hybrid Recommendation Engine, delivered through an interactive Streamlit application.

---

🎯 Project Objective

Design and implement a production-style recommendation system that:

• Learns user preferences
• Finds similar products
• Ranks results intelligently
• Simulates real-world product discovery systems

---

🏗️ System Architecture

The system follows a modular recommendation pipeline:

User Query / Interaction
        ↓
User History Extraction
        ↓
Candidate Generation
   ├─ Content-Based Filtering (TF-IDF)
   └─ Collaborative Filtering (SVD)
        ↓
Hybrid Ranking Engine
        ↓
Top-N Recommendations

---

🔄 Recommendation Pipeline

graph TD
User[User Query] --> History[User Interaction Data]
History --> CF["Collaborative Filtering - SVD"]
User --> CB["Content-Based Filtering - TF-IDF"]
CF --> Hybrid[Hybrid Scoring]
CB --> Hybrid
Hybrid --> Output[Top-N Recommendations]

Loading

---

⚙️ Hybrid Scoring Formula

Final ranking score is computed as:

Final Score =
0.6 × Collaborative Score
+
0.4 × Content Similarity Score

This balances:

• User behavior (collaborative filtering)
• Product similarity (content-based filtering)

---

📊 Dataset

Dataset derived from:

Mercari Price Suggestion Challenge (Kaggle)
https://www.kaggle.com/c/mercari-price-suggestion-challenge

Dataset Characteristics

• Users: ~10,000 (simulated interactions)
• Products: ~2,500+
• Features:
  • Product title
  • Category
  • Description
  • Price

---

⚙️ Data Preparation

Run:

python prepare_products.py

Generates:

data/products.csv
data/interactions.csv

Feature Engineering

• Text preprocessing (cleaning titles/descriptions)
• TF-IDF vectorization of product text
• User-item interaction matrix construction
• Normalisation of product metadata

These steps enable both content similarity and collaborative learning.

---

🤖 Recommendation Models

1️⃣ Content-Based Filtering

• TF-IDF vectorisation
• Cosine similarity
• Captures product similarity

2️⃣ Collaborative Filtering

• SVD (matrix factorization using Surprise library)
• Learns user-item interaction patterns

3️⃣ Hybrid Recommendation

• Combines both approaches
• Produces more accurate and personalized results

---

🔎 Search & Ranking System

The system simulates a real-world search ranking pipeline:

User Query
    ↓
Candidate Retrieval (TF-IDF similarity)
    ↓
Ranking Model (SVD + Hybrid Score)
    ↓
Top-N Results

This mirrors how modern e-commerce systems rank search results.

---

📈 Evaluation Metrics

Recommendation quality can be evaluated using:

Metric	Description
Precision@K	Relevance of top-K recommendations
Recall@K	Coverage of relevant items
MAP	Mean Average Precision
NDCG	Ranking quality

Example (simulated):

Precision@10: 0.82  
Recall@10: 0.74  
NDCG@10: 0.86  

---

🚀 Key Features

🔎 Smart Product Search

Users can search queries such as:

Nike running shoes
Jordan sneakers
Adidas ultraboost

Returns top-N ranked recommendations.

---

🤖 Hybrid Recommendation Engine

• Combines collaborative + content-based filtering
• Improves recommendation accuracy
• Handles cold-start problems

---

🖥️ Interactive Web Interface

• Product search
• Trending recommendations
• Product cards with images
• External purchase links

---

🎯 Advanced Features

• Price filtering
• Category filtering
• Sorting:
  • Best Match
  • Price Low → High
  • Price High → Low

---

📈 Example Workflow

User searches: "Nike running shoes"
        ↓
TF-IDF retrieves similar products
        ↓
SVD ranks based on user interactions
        ↓
Hybrid score computed
        ↓
Top recommendations displayed

---

🧩 Project Structure

marketplace-recommendation-engine
│
├── app
│   └── streamlit_app.py
│
├── src
│   ├── similarity_search.py
│   ├── collaborative_filtering.py
│   ├── hybrid_recommender.py
│   └── image_fetcher.py
│
├── data
│   ├── products.csv
│   └── interactions.csv
│
├── tests
│
├── prepare_products.py
├── requirements.txt
└── README.md

---

⚡ Performance Considerations

To support scalability:

• Precomputed TF-IDF vectors for fast similarity search
• Efficient matrix factorization using SVD
• Batch recommendation generation

Potential improvements:

• Redis caching for frequent queries
• FAISS / vector database for similarity search
• Microservices-based recommendation system
• Distributed model serving

---

▶️ Run Locally

Clone:

git clone https://github.com/premnadh/marketplace-recommendation-engine.git

Navigate:

cd marketplace-recommendation-engine

Create virtual environment:

python3.11 -m venv venv

Activate:

source venv/bin/activate

Install dependencies:

pip install -r requirements.txt

Run application:

streamlit run app/streamlit_app.py

Open:

http://localhost:8501

---

🌟 Key Highlights

• Hybrid recommendation system
• Real machine learning implementation
• Search + ranking pipeline
• Modular architecture
• Interactive UI

---

🚀 Future Improvements

• Deep learning recommendation models
• Real-time personalization
• Session-based recommendations
• Advanced ranking algorithms
• Cloud deployment (AWS / GCP)

---

👤 Author

Prem Nadh Gajula

Aspiring Data Scientist | Machine Learning Engineer | Backend Developer

---

⭐ If you found this project useful, consider starring the repository!# premnadh