F.30. pg_trgm

F.30.1. Trigram (or Trigraph) Concepts

A trigram is a group of three consecutive characters taken from a string. We can measure the similarity of two strings by counting the number of trigrams they share. This simple idea turns out to be very effective for measuring the similarity of words in many natural languages.

Note: pg_trgm ignores non-word characters (non-alphanumerics) when extracting trigrams from a string. Each word is considered to have two spaces prefixed and one space suffixed when determining the set of trigrams contained in the string. For example, the set of trigrams in the string "cat" is " c", " ca", "cat", and "at ". The set of trigrams in the string "foo|bar" is " f", " fo", "foo", "oo ", " b", " ba", "bar", and "ar ".

F.30.2. Functions and Operators

The functions provided by the pg_trgm module are shown in Table F-25, the operators in Table F-26.

F.30.3. Index Support

The pg_trgm module provides GiST and GIN index operator classes that allow you to create an index over a text column for the purpose of very fast similarity searches. These index types support the above-described similarity operators, and additionally support trigram-based index searches for LIKE, ILIKE, ~ and ~* queries. (These indexes do not support equality nor simple comparison operators, so you may need a regular B-tree index too.)

Example:

CREATE TABLE test_trgm (t text);
CREATE INDEX trgm_idx ON test_trgm USING GIST (t gist_trgm_ops);

or

CREATE INDEX trgm_idx ON test_trgm USING GIN (t gin_trgm_ops);

At this point, you will have an index on the t column that you can use for similarity searching. A typical query is

SELECT t, similarity(t, 'word') AS sml
  FROM test_trgm
  WHERE t % 'word'
  ORDER BY sml DESC, t;

This will return all values in the text column that are sufficiently similar to word, sorted from best match to worst. The index will be used to make this a fast operation even over very large data sets.

A variant of the above query is

SELECT t, t <-> 'word' AS dist
  FROM test_trgm
  ORDER BY dist LIMIT 10;

This can be implemented quite efficiently by GiST indexes, but not by GIN indexes. It will usually beat the first formulation when only a small number of the closest matches is wanted.

Beginning in PostgreSQL 9.1, these index types also support index searches for LIKE and ILIKE, for example

SELECT * FROM test_trgm WHERE t LIKE '%foo%bar';

The index search works by extracting trigrams from the search string and then looking these up in the index. The more trigrams in the search string, the more effective the index search is. Unlike B-tree based searches, the search string need not be left-anchored.

Beginning in PostgreSQL 9.3, these index types also support index searches for regular-expression matches (~ and ~* operators), for example

SELECT * FROM test_trgm WHERE t ~ '(foo|bar)';

The index search works by extracting trigrams from the regular expression and then looking these up in the index. The more trigrams that can be extracted from the regular expression, the more effective the index search is. Unlike B-tree based searches, the search string need not be left-anchored.

For both LIKE and regular-expression searches, keep in mind that a pattern with no extractable trigrams will degenerate to a full-index scan.

The choice between GiST and GIN indexing depends on the relative performance characteristics of GiST and GIN, which are discussed elsewhere. As a rule of thumb, a GIN index is faster to search than a GiST index, but slower to build or update; so GIN is better suited for static data and GiST for often-updated data.

F.30.4. Text Search Integration

Trigram matching is a very useful tool when used in conjunction with a full text index. In particular it can help to recognize misspelled input words that will not be matched directly by the full text search mechanism.

The first step is to generate an auxiliary table containing all the unique words in the documents:

CREATE TABLE words AS SELECT word FROM
        ts_stat('SELECT to_tsvector(''simple'', bodytext) FROM documents');

where documents is a table that has a text field bodytext that we wish to search. The reason for using the simple configuration with the to_tsvector function, instead of using a language-specific configuration, is that we want a list of the original (unstemmed) words.

Next, create a trigram index on the word column:

CREATE INDEX words_idx ON words USING GIN (word gin_trgm_ops);

Now, a SELECT query similar to the previous example can be used to suggest spellings for misspelled words in user search terms. A useful extra test is to require that the selected words are also of similar length to the misspelled word.

Note: Since the words table has been generated as a separate, static table, it will need to be periodically regenerated so that it remains reasonably up-to-date with the document collection. Keeping it exactly current is usually unnecessary.

F.30.5. References

GiST Development Site http://www.sai.msu.su/~megera/postgres/gist/

Tsearch2 Development Site http://www.sai.msu.su/~megera/postgres/gist/tsearch/V2/

F.30.6. Authors

Oleg Bartunov <[email protected]>, Moscow, Moscow University, Russia

Teodor Sigaev <[email protected]>, Moscow, Delta-Soft Ltd.,Russia

Documentation: Christopher Kings-Lynne

This module is sponsored by Delta-Soft Ltd., Moscow, Russia.