Graph-Based Personalized E-commerce Recommendation Engine

Project Overview

This project focuses on building a Graph-based recommender system that recommends similar products to users of an e-commerce platform. It's a continuation of a previous project where we converted clickstream data into various types of graphs.

Business Objective

To create a recommendation system for our eCommerce platform by converting generated graphs into embeddings and using similarity search techniques.

Tech Stack

• Language: Python
• Packages: pandas, numpy, pecanpy, gensim, plotly, umap, faiss
• File Management: Parquet

Approach

1. Understand the problem statement
2. Deepwalk and Node2vec Model training
   • Import necessary packages and libraries
   • Read the selected graph
   • Save the graph in .edg format
   • Graph random walk generation
   • Model Building with Word2Vec and Node2Vec
   • Save the embeddings in parquet format
3. Result analysis and Visualization
   • Import necessary packages and libraries
   • Read the data
   • Define a function for generating levels of category
   • Use UMAP for dimension reduction
   • Visualize the clusters
4. Embedding vector search
   • Import necessary packages and libraries
   • Use FAISS for generating recommendations

Key Achievements

• Successfully converted complex clickstream data into meaningful graph representations
• Implemented advanced embedding techniques like Deepwalk and Node2vec
• Utilized FAISS for efficient similarity search, enabling fast and accurate recommendations
• Created visualizations using UMAP to understand the clustering of products