Technical GlossaryNatural Language Processing

Reranking

A second-stage quality-improvement method that reranks candidates from the first retrieval stage using a stronger model.

Reranking is used in retrieval systems to balance speed and quality in a layered way. The first stage retrieves fast but coarse candidates, and the second stage refines the ranking using a more expensive but more accurate model. This can make a major quality difference in QA, semantic search, and RAG systems. It shows that retrieval is not just about finding candidates, but about ranking them correctly.