Query expansion
Query expansion is rewriting a search query into one or more variants, with synonyms, related terms or different phrasing, to find relevant pages that the original wording would miss. The variants are run as separate searches or merged into one query, and the results are combined. Query expansion raises recall but can lower precision when an added term drifts from the original meaning.
It goes back to relevance feedback in the SMART system, described by Rocchio in 1971, and to pseudo-relevance feedback, which adds terms from the top results of a first search.
Example
For the question “how to stop a laptop from overheating”, an agent using query expansion also searches for “laptop thermal throttling fix”, “clean laptop fan” and “undervolt laptop CPU”. The second and third queries find repair guides that never use the word “overheating”.
FAQ
Does query expansion always improve results?
No. Query expansion helps when the original wording is narrow or unusual, but an added term that drifts from the intended meaning can bring in irrelevant pages, so expanded queries are usually checked or weighted.
How do AI agents use query expansion?
An AI agent uses query expansion by asking a language model for several rewrites of a question, searching with each one and merging the results before it reads the most promising pages.
Related terms
Written by Keenable.
Published
