Definition

Semantic expansion (often called query expansion) adds meaning-aware alternatives to a query, such as synonyms, related concepts, or entity variants. The goal is to retrieve relevant items that do not contain the exact words the user typed.

Why it matters

Higher recall: captures wording variation across documents and languages.
Better domain coverage: maps abbreviations, citations, and jargon to standard forms.
Improved user experience: fewer “no results” searches and fewer reformulations.
Works with hybrid search: expansion can feed both full-text and vector retrieval.

How it works

Query -> understand meaning -> expand safely -> retrieve -> rank -> validate with analytics

Expansion sources include curated synonym lists, taxonomies, and embedding-based nearest neighbors. The hard part is avoiding expansions that change intent.

Practical example

If users search “CIR92”, expansion can include “WIB92” and “Code des impôts sur les revenus” so the same legal source is found across language variants.

Common questions

Q: Why can expansion hurt relevance?

A: Over-expansion adds noise. Good systems expand selectively, weight expansions lower, and monitor outcomes.

Q: Is this the same as synonyms in the index?

A: Related, but not identical. You can expand at query time, at index time, or both. Query-time expansion is usually easier to audit and adjust.

Query Understanding - interpret meaning before expanding
Query Intent - keep expansions aligned with the user goal
Embeddings - power meaning-based expansion
Full-Text Search - expansion can add keyword variants
Relevance Tuning - measure and refine expansion impact

References

Manning, Raghavan & Schütze (2008), Introduction to Information Retrieval.