ConceptReranking / 文档重排 / 重排模块
Reranking
A module in a RAG pipeline that performs secondary sorting of candidate documents retrieved in an initial retrieval stage. It is typically implemented using a cross-encoder and achieves higher precision than vector similarity-based retrieval.
Timeline (last 90 days)
Oct 28
Vector retrieval (coarse ranking) is fast but lacks deep semantic understanding and may recall irrelevant content, so re-ranking is needed for fine ranking.
Unverified50%