Glossary · E-commerce ML

Learning to Rank

also: LTR · L2R · search ranking

Definition

Learning to Rank is the class of supervised machine learning algorithms that optimize the ordering of a result set, search results, recommendations, product rankings, for revenue, engagement, or relevance. Pairwise (RankNet) and listwise (LambdaMART, ListNet) objectives are the dominant training paradigms.

Classical information retrieval ranked documents by hand-tuned relevance scores (TF-IDF, BM25). Learning to Rank replaces the hand-tuning with supervised learning over labeled judgments or implicit click data. Pointwise approaches predict a per-item score; pairwise (RankNet) optimize pairs of items; listwise (LambdaMART, ListNet) optimize the entire ranked list against metrics like NDCG. In e-commerce, revenue-optimized ranking replaces relevance-only objectives with composite utility functions including margin, inventory, and long-term user value.

Essays on this concept