Trigram Similarity Search in MySQL

VillageSQL is a drop-in replacement for MySQL with extensions.

All examples in this guide work on VillageSQL. Install Now →

This guide requires VillageSQL 0.0.4 or later.

MySQL’s built-in string functions — LIKE, SOUNDEX, LEVENSHTEIN — cover basic matching, but none of them give you a continuous similarity score you can sort or threshold on. VillageSQL’s vsql_trgm extension adds trigram-based similarity: a number between 0 and 1 that measures how alike two strings are, so you can rank results and filter by similarity in a single query.

How trigrams work

A trigram is a group of 3 consecutive characters. vsql_trgm lowercases the input, splits it on non-alphanumeric characters, pads each word with two leading spaces and one trailing space, then slides a 3-character window across each padded word. Similarity is how many trigrams the two inputs share, divided by the larger of the two trigram set sizes.

"cat" → pad to "  cat " → ["  c", " ca", "cat", "at "]
"car" → pad to "  car " → ["  c", " ca", "car", "ar "]
shared: 2 of 4 → similarity 0.5

Two things to know about multi-word strings: they are lowercased before processing, so "Cat" and "cat" produce the same trigrams. They are also split on non-alphanumeric characters, so each word is padded and trigrammed independently — "hello world" produces trigrams from " hello " and " world " separately, not from a single padded string. There is no "o w" or " wo" trigram. Similarity ranges from 0 (nothing in common) to 1 (identical).

Getting started

INSTALL EXTENSION vsql_trgm;

Core functions

trgm_similarity — continuous score

SELECT vsql_trgm.trgm_similarity('adidas', 'adidos');   -- 0.5714...
SELECT vsql_trgm.trgm_similarity('cat', 'dog');          -- 0
SELECT vsql_trgm.trgm_similarity('mysql', 'mysql');      -- 1

All functions return NULL when any argument is NULL.

trgm_similar — threshold check

Returns 1 if similarity is at or above 0.3 (the PostgreSQL default), 0 otherwise:

SELECT vsql_trgm.trgm_similar('cat', 'car');   -- 1
SELECT vsql_trgm.trgm_similar('cat', 'dog');   -- 0

For a custom threshold:

SELECT vsql_trgm.trgm_similar_threshold('cat', 'car', 0.6);  -- 0
SELECT vsql_trgm.trgm_similar_threshold('cat', 'car', 0.4);  -- 1

trgm_distance — complement of similarity

1 - similarity. Useful for ORDER BY when you want closest first:

SELECT vsql_trgm.trgm_distance('mysql', 'mysqk');  -- ~0.33

Finding near-matches in a table

The most common use: find rows that resemble a search term.

SELECT
    name,
    vsql_trgm.trgm_similarity(name, 'adidas') AS score
FROM products
WHERE vsql_trgm.trgm_similarity(name, 'adidas') > 0.3
ORDER BY score DESC
LIMIT 10;

This handles typos, alternate spellings, and partial matches — cases where LIKE '%adidas%' would miss.

Word similarity

trgm_word_similarity finds the best match between a query string and any contiguous substring of the target. This is useful when the search term might be a word inside a longer string.

-- Find products whose name contains something close to "widget"
SELECT name, vsql_trgm.trgm_word_similarity('widget', name) AS score
FROM products
WHERE vsql_trgm.trgm_word_similar('widget', name) = 1
ORDER BY score DESC;

trgm_strict_word_similarity is a stricter variant that requires the match to align with word boundaries.

Function	Threshold
`trgm_word_similar(a, b)`	0.6
`trgm_strict_word_similar(a, b)`	0.5

Inspecting trigrams

trgm_show returns the sorted trigram set for a string as a JSON array:

SELECT vsql_trgm.trgm_show('cat');
-- '["  c"," ca","at ","cat"]'

This is useful for debugging why two strings score the way they do.

Function reference

Function	Returns	Notes
`trgm_similarity(a, b)`	REAL	0–1 similarity score
`trgm_distance(a, b)`	REAL	1 − similarity
`trgm_similar(a, b)`	INT	1 if similarity ≥ 0.3
`trgm_similar_threshold(a, b, t)`	INT	1 if similarity ≥ t
`trgm_word_similarity(a, b)`	REAL	best substring match score
`trgm_word_distance(a, b)`	REAL	1 − word_similarity
`trgm_word_similar(a, b)`	INT	1 if word_similarity ≥ 0.6
`trgm_strict_word_similarity(a, b)`	REAL	word-boundary-aligned score
`trgm_strict_word_distance(a, b)`	REAL	1 − strict_word_similarity
`trgm_strict_word_similar(a, b)`	INT	1 if strict_word_similarity ≥ 0.5
`trgm_show(text)`	TEXT	JSON array of trigrams

Known limitations

No index acceleration. VEF does not expose a custom index API, so WHERE trgm_similarity(col, 'term') > 0.3 requires a full table scan. For large tables, pre-filter with LIKE or full-text search to reduce the candidate set, then re-score with trgm_similarity. No session-level threshold. PostgreSQL’s set_limit() / show_limit() cannot be implemented in VEF (VEF functions are stateless). Use trgm_similar_threshold(a, b, t) to pass the threshold explicitly. No array return. trgm_show returns a JSON string rather than a native array type.

Getting Started

Server Development

Extensions

Using Extensions

Creating Extensions

VillageSQL is a drop-in replacement for MySQL with extensions.

How trigrams work

Getting started

Core functions

trgm_similarity — continuous score

trgm_similar — threshold check

trgm_distance — complement of similarity

Finding near-matches in a table

Word similarity

Inspecting trigrams

Function reference

Known limitations

See also

VillageSQL is a drop-in replacement for MySQL with extensions.

​How trigrams work

​Getting started

​Core functions

​trgm_similarity — continuous score

​trgm_similar — threshold check

​trgm_distance — complement of similarity

​Finding near-matches in a table

​Word similarity

​Inspecting trigrams

​Function reference

​Known limitations

​See also

How trigrams work

Getting started

Core functions

trgm_similarity — continuous score

trgm_similar — threshold check

trgm_distance — complement of similarity

Finding near-matches in a table

Word similarity

Inspecting trigrams

Function reference

Known limitations

See also