To do fuzzy searching with PostgreSQL, you can add the pg_trgm extension. To activate the extension, just run the following instruction:
test=# CREATE EXTENSION pg_trgm; CREATE EXTENSION
The pg_trgm extension is pretty powerful, and to show what it is capable of, I have compiled some sample data consisting of 2,354 names of villages and cities here in Austria, Europe.
My company website has all the data and PostgreSQL enables you to load the data directly:
test=# COPY t_location FROM PROGRAM 'curl www.cybertec.at/secret/orte.txt'; COPY 2354
Note that curl (a command-line tool to fetch data) has to be installed. If you don't have this tool, download the file normally and import it from your local filesystem.
Once the data has been loaded, it is possible to check out the content of the table:
test=# SELECT * FROM t_location LIMIT 4; name -------------------------------- Eisenstadt Rust Breitenbrunn am Neusiedler See Donnerskirchen (4 rows)
If German is not your mother tongue, it will be impossible to spell the names of those locations without severe mistakes. Fortunately, pg_trgm will come to the rescue:
test=# CREATE EXTENSION pg_trgm; CREATE EXTENSION
The pg_trgm provides us with a distance operator, which computes the distance between two strings:
Those sequences will then be used to come up with the distance you have just seen. Of course, the distance operator can be used inside a query to find the closest match:
test=# SELECT * FROM t_location ORDER BY name <-> 'Kramertneusiedel' LIMIT 3; name ----------------- Gramatneusiedl Klein-Neusiedl Potzneusiedl (3 rows)
Gramatneusiedl is pretty close to Kramertneusiedel. It sounds similar and using a K instead of a G is a pretty common mistake. On Google, you will sometimes see Did you mean.... It is quite likely that Google is using n-grams here to do that.
In PostgreSQL, it is possible to use GiST to index on text using trigrams:
test=# CREATE INDEX idx_trgm ON t_location USING GiST(name GiST_trgm_ops); CREATE INDEX
pg_trgm provides us with the GiST_trgm_ops operator class designed to do similarity searches. The following listing shows that the index is used as expected:
test=# explain SELECT * FROM t_location ORDER BY name <-> 'Kramertneusiedel' LIMIT 5; QUERY PLAN ----------------------------------------------------------------- Limit (cost=0.14..0.58 rows=5 width=17) -> Index Scan using idx_trgm on t_location (cost=0.14..207.22 rows=2354 width=17) Order By: (name <-> 'Kramertneusiedel'::text) (3 rows)