Knowledge Maxxing← Back to topic library
AI MaxxingIntermediate75 sec
Reranking
Reranking is a second pass over an initial set of retrieved results that reorders them by a more precise relevance score before they are used or shown.
Prerequisite: Information retrieval
What matters
Initial retrieval is usually optimized for speed over a huge collection, so a reranker can afford a slower, more accurate comparison over just the smaller shortlist it returned.
Example
A search system might retrieve 100 candidate documents quickly, then rerank just those 100 with a slower, more accurate model to pick the best 5.
Common misconception
Reranking does not replace retrieval; it only reorders what retrieval already found, so a poor initial retrieval still limits the final result.
Recall question
Why can a reranker afford to be slower than the initial retrieval step?
Keywords and related concepts
rerankingrelevance scoringretrievalretrieval-augmented generation