{"id":4132,"date":"2017-10-31T08:30:10","date_gmt":"2017-10-31T15:30:10","guid":{"rendered":"https:\/\/www.couchbase.com\/blog\/?p=4132"},"modified":"2025-06-13T20:59:29","modified_gmt":"2025-06-14T03:59:29","slug":"query-natural-language-couchbase-servers-powerful-full-text-search","status":"publish","type":"post","link":"https:\/\/www.couchbase.com\/blog\/query-natural-language-couchbase-servers-powerful-full-text-search\/","title":{"rendered":"Query Natural Language with Couchbase Server’s Powerful Full Text Search"},"content":{"rendered":"

How many times have you written a SQL query that used a LIKE<\/code> operator and some wildcards to find text in a string? What kind of performance did you get when you ran that against millions of records? It was probably not good, right?<\/p>\n

While you could use wildcards in SQL, it probably isn’t the best way for most scenarios. This includes N1QL queries. Instead, meet Full Text Search (FTS), a technology that has been around for a while, but recently included in Couchbase<\/a> Server.<\/p>\n

<\/p>\n

The Full Text Search found in Couchbase is based on Bleve<\/a>, a search and indexing library written in Golang.<\/p>\n

Fuzzy Searching with Full Text Search in Couchbase<\/h2>\n

When comparing FTS against a wildcard N1QL or SQL query, there is more to compare against than just the performance aspect. Take for example, its ability to match in a fuzzy fashion. Let’s assume we had the following query:<\/p>\n

SELECT * FROM `default` WHERE message LIKE '%bananas%';<\/pre>\n
What happens if the database has a record containing Bananas<\/strong> with a capital B<\/strong>? In this circumstance it would not be included in the results. What happens if you’re like me and can’t spell, leaving results like\u00a0bannanas<\/strong> in the database? It will not be included in the results either.<\/p>\n
With Full Text Search, the following query can be done with a fuzziness factor:<\/p>\n
banana~2<\/pre>\n
Ignoring that SQL and FTS have a different query syntax, the above would offer two factors of fuzziness. This means that two characters in the search term can be altered to get to our result. Our data could include any of the following:<\/p>\n
banana\r\nbannanna\r\nbana\r\nbnan<\/pre>\n
It isn’t limited to just the four above, but you get the idea.<\/p>\n
However, what if our database contains\u00a0bandana<\/strong> which in this case isn’t a typing mistake. Do we really want searches for legitimate other items appearing in our results?<\/p>\n
When executing a Full Text Search query for a term or set of terms, each result is scored based on how relevant it is to the initial search query. You can wrap your business logic around the scored results.<\/p>\n
The Different Types of Queries in Couchbase’s Full Text Search<\/h2>\n
There are many different types of FTS queries<\/a> that can be performed in Couchbase. The type of query I’ve been mentioning so far is best classified as a Match Query, where the search term is used to match against the index with and without fuzziness.<\/p>\n
A few other types of queries include, but are not limited to:<\/p>\n