Products Purchased by Those Who Viewed This Item

This paper presents two recommendation models for e-commerce platforms, enhancing product discovery by suggesting alternatives based on historical purchasing patterns associated with product views. The first model employs a statistical approach, while the second utilizes a neural network-based two-tower architecture. Both models adhere to the "Products Purchased by Those Who Viewed This Item" principle. Extensive online A/B testing revealed that the statistical approach significantly improved click-through rates (CTR) by 2.55% and conversion rates (CR) by 0.51% compared to conventional similarity-based recommendation methods. However, the two-tower model, despite offering broader coverage, resulted in a 1% decrease in CTR and a 0.21% decrease in CR compared to the statistical approach. These findings underscore the potential of sophisticated modeling techniques to apply user behavior data, significantly enhancing user engagement and profitability in e-commerce settings. Consequently, the two-tower model is implemented as a supplementary recommendation system to enrich the user experience alongside the statistical approach.