Web Dev//SQL//fuzzy matching

Text search that tolerates typos, partial matches, and misspellings: returns results ranked by similarity instead of requiring exact matches.

Text search that tolerates typos, partial matches, and misspellings: returns results ranked by similarity instead of requiring exact matches.

"atention" matches "Attention", "transform" matches "transformers". The query doesn't need to be perfect

Works by comparing character n-grams (trigrams) or edit distance (Levenshtein) between the query and the stored text

Returns a similarity score (0.0–1.0) so results can be ranked by relevance, not just filtered yes/no

The % operator in PostgreSQL (via pg_trgm) returns true if similarity exceeds a threshold. Combine with ORDER BY similarity() for ranked results

Contrast with substring matching (LIKE '%term%'): substring is exact and unranked: either the string contains the term or it doesn't. Fuzzy matching handles human error.

Contrast with full-text search (tsvector/tsquery): FTS tokenizes and stems words, understands language structure. Fuzzy matching is character-level, dumber but catches typos that FTS misses. They complement each other.