Tutorials & Tips

Introducing contextual retrieval: a new method that greatly improves RAG accuracy

Original title:Introducing Contextual Retrieval

AI summary

Traditional RAG systems easily lose context when encoding, leading to retrieval failures. The new contextual retrieval method uses two sub-techniques, contextual embeddings and contextual BM25, to prepend explanatory context to chunks before retrieval, reducing retrieval failures by 49% and by up to 67% when combined with reranking, significantly improving downstream task performance. For small knowledge bases under 200K tokens, the full content can be placed directly in the prompt; combined with Claude's prompt caching, this cuts latency by more than 2x and costs by up to 90%. For large knowledge bases, contextual retrieval offers a scalable solution.

Key Takeaways

  • Traditional RAG systems easily lose context when encoding,leading to retrieval failures.
  • The new contextual retrieval method uses two sub-techniques,contextual embeddings and contextual…
  • For small knowledge bases under 200K tokens,the full content can be placed directly in the prompt
  • combined with Claude's prompt caching,this cuts latency by more than 2x and costs by up to 90%.
  • For large knowledge bases,contextual retrieval offers a scalable solution.

Quick Facts

Published
Updated
Source
Anthropic:Engineering(事故复盘 + 工程实践 · 网页)
Category
Tutorials & Tips
AI score
72
Read the original Read the full article at the original source
Why it matters

Anthropic 把 RAG 检索失败率砍了 67%,方法不复杂但数据扎实,做知识库的开发者可以直接抄 cookbook 上手,是那种读完当天就能改进生产环境的实用帖。

Related entities

Source & copyright

This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.

Original article · AIHOT entry

How to Cite

When citing this article, keep the title, date, and permalink, and credit Orange News as the source.

Orange News: Introducing contextual retrieval: a new method that greatly improves RAG accuracy. September 19, 2024. https://new.nat6.net/en/news/cmorb7ml1008tslhfm0c04uo4/

View all →