Item Cold Start in E-Commerce Recommender Systems: A Survey

Recommender systems in e-commerce face a persistent challenge known as the item cold-start problem, where systems struggle to make relevant suggestions about new items due to insufficient historical data. While early research primarily focused on effectiveness metrics like accuracy, recent years have seen a change toward incorporating fairness and explainability into the systems. The evolution of transformer architectures has significantly impacted data complexity. This also marks a shift enabling systems to power recommendations on complex patterns in data, but with a requirement of larger datasets. This survey categorizes existing approaches to the item cold-start problem using a three-tier classification system that distinguishes methodological foundation, data structure and implementation technique and presents them in a temporal framework that highlights both data complexity and underlying objectives. The survey outlines a practical framework tailored to different e-commerce business scales, guiding implementation strategies accordingly. Additionally, evaluation metrics are analyzed through the lens of effectiveness and fairness, providing practitioners with a comprehensive guide for implementing these systems in practice.