Skip to main content
All terms
Patterns

Reranking

Reordering retrieved candidates with a stronger model to surface the most relevant first.

Definition

Reranking takes an initial set of retrieved candidates and reorders them with a stronger, slower model so the most relevant items rise to the top. It is a common second stage in retrieval pipelines: a fast search returns many candidates, and the reranker scores each against the query more carefully. This improves the quality of the context fed to a model without scoring every document in the whole collection.