Query Expansion
Definition
Query expansion augments the original user query with synonyms, related terms, or LLM-generated variations to improve retrieval recall and find relevant documents the original query might miss.
Why It Matters
Users often don’t phrase queries optimally for retrieval. They might use different terms than the documents, ask vague questions, or miss important keywords. Query expansion bridges this vocabulary gap, significantly improving recall without changing your document collection.
Common Techniques
LLM-based Expansion:
- Generate synonyms and related terms
- Rephrase the query in multiple ways
- Create hypothetical answers (HyDE)
- Break complex queries into sub-queries
Multi-Query RAG:
- Generate multiple query variants
- Retrieve for each variant
- Combine and deduplicate results
When to Use
Use query expansion when: retrieval recall is low, users phrase things differently than your documents, queries are often vague or underspecified, or you’re seeing relevant documents missed by direct queries. Balance against the added latency and cost of multiple retrievals.