Unverified90% confidenceFactTime unknown

重排（Reranking）通常使用交叉编码器对每个片段与用户问题的相关性重新评估

1

Sources

90%

Confidence

Long-term

Relevance

6/3/2026

First Seen

Sources

RAG System End-to-End Breakdown: From Vector Indexing to Production Optimization

bilibili大模型本地部署

Related Entities

Rerankingconcept Cross-Encoderconcept

Related Claims

UnverifiedThe reranking module typically uses a cross-encoder model to compute relevance scores between queries and documents, performing a secondary ranking of the candidate documents from the initial retrieval.77% similar Unverified该开发者的Agentic代码审查流程采用生成→审查→修补→再审查的循环迭代，直到代码通过质量门槛62% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/38318

API

curl https://kongchang.com/api/v1/knowledge/claims/38318

MCP

get_claim(id=38318)