Comparative Evaluation of Word2Vec and Node2Vec for Frequently Bought Together Recommendations in E-Commerce

In this study, we conducted a comparative evaluation of two popular embedding techniques, Word2Vec and Node2Vec, to enhance recommendation systems for frequently bought together products in an e-commerce setting. Utilizing historical purchase data, we generated product embeddings and assessed their performance through both offline e valuation and online A/B testing. Our results demonstrate that Node2Vec outperforms Word2Vec in terms of click-through rates (CTR) and conversion rates (CR), achieving relative improvements of 8.96% and 6.04%, respectively. This paper details the methodologies applied, the experimental setup, and the implications of using graph-based embedding techniques over traditional vector-based methods. By comparing these approaches, we provide insights into how different embedding strategies can significantly impact the effectiveness of recommendation systems in real-world e-commerce environments.