Definition
The process of rewriting or expanding a user's initial input into one or more optimized versions to improve the precision and recall of the retrieval step in a RAG pipeline. While it significantly enhances search relevance by resolving ambiguities and injecting context, it introduces additional latency and token costs due to the required intermediate LLM call.
Distinguish from 'Query Expansion' by its focus on transforming the semantic structure rather than just adding synonyms.
"A skilled translator taking a vague, slang-heavy request and turning it into a formal, precise library catalog search string."
- HyDE (Hypothetical Document Embeddings)(Specialized Technique)
- Multi-Query Retrieval(Implementation Strategy)
- Contextual Retrieval(Prerequisite)
- Step-back Prompting(Reasoning Strategy)
Conceptual Overview
The process of rewriting or expanding a user's initial input into one or more optimized versions to improve the precision and recall of the retrieval step in a RAG pipeline. While it significantly enhances search relevance by resolving ambiguities and injecting context, it introduces additional latency and token costs due to the required intermediate LLM call.
Disambiguation
Distinguish from 'Query Expansion' by its focus on transforming the semantic structure rather than just adding synonyms.
Visual Analog
A skilled translator taking a vague, slang-heavy request and turning it into a formal, precise library catalog search string.