What is a ranking model in recommendation systems?

A ranking model orders candidate items based on predicted relevance to a user.

Ranking models are used after candidate generation to sort a limited set of items (hundreds or thousands) by predicted user interest. These models use features such as user history, item metadata, context (time, device), and embeddings. Modern ranking systems often use Gradient Boosted Decision Trees or deep learning models. The goal is to optimize ranking metrics like NDCG or CTR rather than raw prediction accuracy.