Building a Recommendation System Using Amazon Product Co-Purchasing Network

TL;DR

This paper presents an online, inductive recommendation system for new products on e-commerce platforms, utilizing a co-purchasing graph and a modified GraphSAGE method to predict relevant items in real time.

Contribution

It introduces a scalable, real-time recommendation approach for new products using a graph neural network that generalizes to unseen items.

Findings

Outperforms baseline algorithms in link prediction accuracy

Effectively recommends new products with limited initial information

Scales to large, evolving product catalogs

Abstract

This project develops an online, inductive recommendation system for newly listed products on e-commerce platforms, focusing on suggesting relevant new items to customers as they purchase other products. Using the Amazon Product Co-Purchasing Network Metadata dataset, we construct a co-purchasing graph where nodes represent products and edges capture co-purchasing relationships. To address the challenge of recommending new products with limited information, we apply a modified GraphSAGE method for link prediction. This inductive approach leverages both product features and the existing co-purchasing graph structure to predict potential co-purchasing relationships, enabling the model to generalize to unseen products. As an online method, it updates in real time, making it scalable and adaptive to evolving product catalogs. Experimental results demonstrate that our approach outperforms baseline algorithms in predicting relevant product links, offering a promising solution for enhancing the relevance of new product recommendations in e-commerce environments. All code is available at https://github.com/cse416a-fl24/final-project-l-minghao_z-catherine_z-nathan.git.