Spelling correction identifies misspelled query terms and generates replacements that are likely to be conventional within a search domain. Search engines can derive candidates from document terms, indexed fields, dictionaries, or precomputed synonym lists, but frequency alone may ignore user context. section 6.5.1
A query-signal approach learns corrections from actual searches. It normalizes queries, tokenizes them when appropriate, removes stop words, short terms, and standalone numbers, then counts valid terms. Infrequent terms become misspelling candidates, while frequent terms become possible canonical spellings. section 6.5.2
A practical correction pipeline uses these steps:
- Calculate frequency quantiles instead of fixed thresholds, because collections vary in size. section 6.5.2
Choosing modem for moden
A product collection returns modes, model, modern, and modem for the query moden. Document frequency alone ranks the generic alternatives, while domain context makes modem the useful correction. User-query analysis produces the mapping moden → modem, alongside examples such as latop → laptop and tosheba → toshiba.
