Reranking
Second-stage ordering of retrieved candidates by a more accurate scoring model.
Overview
Reranking takes the candidate documents returned by a fast first-stage retrieval and reorders them with a slower, more accurate model that reads the query and document together. It solves the accuracy ceiling of vector search, which compares query and document independently and therefore misses fine-grained relevance. It exists because the two goals conflict: retrieval must scan millions of items cheaply, while relevance judgement benefits from reading the pair jointly. Reranking lets each stage do what it is good at.
Technology Graph
How Reranking connects to the rest of the stack.
Uses and depends on
- Cross-Encoder concept Techniques Reranking uses Cross-Encoder High ✓ source
- BERT model Models Reranking based on BERT High ✓ source
Used by
- AI Search application Applications Reranking used by AI Search High ✓ source
- RAG technique Techniques Reranking used by RAG Medium ✓ source