Ranking Metrics: Precision@K, Recall@K, and NDCG
Explore how to evaluate ranked recommendation lists using Precision@K, Recall@K, and NDCG metrics. Understand their differences and how each metric reflects the relevance and order of items in recommender systems. This lesson helps you assess models that deliver ordered recommendations, improve user experience, and make informed choices about metric selection based on product context.
A recommender that places course_17 first and course_42 third can create a different experience from one that reverses those positions, even when both lists contain the same courses. Classification accuracy cannot capture that difference because a recommender produces an ordered list and users often see only its first few results.
We evaluate a ranked list with held-out interactions, relevant items removed from a user’s known history before recommendations are generated. For each user, the recommender returns unseen item_id values, and we check whether the held-out relevant items appear in the first