Full Text Search

How to use full text search in PostgreSQL.

Postgres has built-in functions to handle Full Text Search queries. This is like a "search engine" within Postgres.

Preparation#

For this guide we'll use the following example data:

Usage#

The functions we'll cover in this guide are:

to_tsvector()#

Converts your data into searchable tokens. to_tsvector() stands for "to text search vector." For example:

1
2
select to_tsvector('green eggs and ham');-- Returns 'egg':2 'green':1 'ham':4

Collectively these tokens are called a "document" which Postgres can use for comparisons.

to_tsquery()#

Converts a query string into tokens to match. to_tsquery() stands for "to text search query."

This conversion step is important because we will want to "fuzzy match" on keywords. For example if a user searches for eggs, and a column has the value egg, we probably still want to return a match.

Postgres provides several functions to create tsquery objects:

• to_tsquery() - Requires manual specification of operators (&, |, !)
• plainto_tsquery() - Converts plain text to an AND query: plainto_tsquery('english', 'fat rats') → 'fat' & 'rat'
• phraseto_tsquery() - Creates phrase queries: phraseto_tsquery('english', 'fat rats') → 'fat' <-> 'rat'
• websearch_to_tsquery() - Supports web search syntax with quotes, "or", and negation

Match: @@#

The @@ symbol is the "match" symbol for Full Text Search. It returns any matches between a to_tsvector result and a to_tsquery result.

Take the following example:

1
2
3
select *from bookswhere title = 'Harry';

The equality symbol above (=) is very "strict" on what it matches. In a full text search context, we might want to find all "Harry Potter" books and so we can rewrite the example above:

1
2
3
select *from bookswhere to_tsvector(title) @@ to_tsquery('Harry');

Basic full text queries#

Search a single column#

To find all books where the description contain the word big:

1
2
3
4
5
6
7
select  *from  bookswhere  to_tsvector(description)  @@ to_tsquery('big');

Search multiple columns#

Right now there is no direct way to use JavaScript or Dart to search through multiple columns but you can do it by creating computed columns on the database.

To find all books where description or title contain the word little:

1
2
3
4
5
6
7
select  *from  bookswhere  to_tsvector(description || ' ' || title) -- concat columns, but be sure to include a space to separate them!  @@ to_tsquery('little');

Match all search words#

To find all books where description contains BOTH of the words little and big, we can use the & symbol:

1
2
3
4
5
6
7
select  *from  bookswhere  to_tsvector(description)  @@ to_tsquery('little & big'); -- use & for AND in the search query

Match any search words#

To find all books where description contain ANY of the words little or big, use the | symbol:

1
2
3
4
5
6
7
select  *from  bookswhere  to_tsvector(description)  @@ to_tsquery('little | big'); -- use | for OR in the search query

Notice how searching for big includes results with the word bigger (or biggest, etc).

Partial search#

Partial search is particularly useful when you want to find matches on substrings within your data.

Implementing partial search#

You can use the :* syntax with to_tsquery(). Here's an example that searches for any book titles beginning with "Lit":

1
select title from books where to_tsvector(title) @@ to_tsquery('Lit:*');

Extending functionality with RPC#

To make the partial search functionality accessible through the API, you can wrap the search logic in a stored procedure.

After creating this function, you can invoke it from your application using the SDK for your platform. Here's an example:

1
2
3
4
5
6
7
create or replace function search_books_by_title_prefix(prefix text)returns setof books AS $$begin  return query  select * from books where to_tsvector('english', title) @@ to_tsquery(prefix || ':*');end;$$ language plpgsql;

This function takes a prefix parameter and returns all books where the title contains a word starting with that prefix. The :* operator is used to denote a prefix match in the to_tsquery() function.

Handling spaces in queries#

When you want the search term to include a phrase or multiple words, you can concatenate words using a + as a placeholder for space:

1
select * from search_books_by_title_prefix('Little+Puppy');

Web search syntax with websearch_to_tsquery()#

The websearch_to_tsquery() function provides an intuitive search syntax similar to popular web search engines, making it ideal for user-facing search interfaces.

Basic usage#

1
2
3
select *from bookswhere to_tsvector(description) @@ websearch_to_tsquery('english', 'green eggs');

Quoted phrases#

Use quotes to search for exact phrases:

1
2
3
select * from bookswhere to_tsvector(description || ' ' || title) @@ websearch_to_tsquery('english', '"Green Eggs"');-- Matches documents containing "Green" immediately followed by "Eggs"

OR searches#

Use "or" (case-insensitive) to search for multiple terms:

1
2
3
select * from bookswhere to_tsvector(description) @@ websearch_to_tsquery('english', 'puppy or rabbit');-- Matches documents containing either "puppy" OR "rabbit"

Negation#

Use a dash (-) to exclude terms:

1
2
3
select * from bookswhere to_tsvector(description) @@ websearch_to_tsquery('english', 'animal -rabbit');-- Matches documents containing "animal" but NOT "rabbit"

Complex queries#

Combine multiple operators for sophisticated searches:

1
2
3
4
select * from bookswhere to_tsvector(description || ' ' || title) @@  websearch_to_tsquery('english', '"Harry Potter" or "Dr. Seuss" -vegetables');-- Matches books by "Harry Potter" or "Dr. Seuss" but excludes those mentioning vegetables

Creating indexes#

Now that you have Full Text Search working, create an index. This allows Postgres to "build" the documents preemptively so that they don't need to be created at the time we execute the query. This will make our queries much faster.

Searchable columns#

Let's create a new column fts inside the books table to store the searchable index of the title and description columns.

We can use a special feature of Postgres called Generated Columns to ensure that the index is updated any time the values in the title and description columns change.

1
2
3
4
5
6
7
8
9
alter table  booksadd column  fts tsvector generated always as (to_tsvector('english', description || ' ' || title)) stored;create index books_fts on books using gin (fts); -- generate the indexselect id, ftsfrom books;

Search using the new column#

Now that we've created and populated our index, we can search it using the same techniques as before:

1
2
3
4
5
6
select  *from  bookswhere  fts @@ to_tsquery('little & big');

Query operators#

Visit Postgres: Text Search Functions and Operators to learn about additional query operators you can use to do more advanced full text queries, such as:

Proximity: <->#

The proximity symbol is useful for searching for terms that are a certain "distance" apart. For example, to find the phrase big dreams, where the a match for "big" is followed immediately by a match for "dreams":

1
2
3
4
5
6
select  *from  bookswhere  to_tsvector(description) @@ to_tsquery('big <-> dreams');

We can also use the <-> to find words within a certain distance of each other. For example to find year and school within 2 words of each other:

1
2
3
4
5
6
select  *from  bookswhere  to_tsvector(description) @@ to_tsquery('year <2> school');

Negation: !#

The negation symbol can be used to find phrases which don't contain a search term. For example, to find records that have the word big but not little:

1
2
3
4
5
6
select  *from  bookswhere  to_tsvector(description) @@ to_tsquery('big & !little');

Ranking search results #

Postgres provides ranking functions to sort search results by relevance, helping you present the most relevant matches first. Since ranking functions need to be computed server-side, use RPC functions and generated columns.

Creating a search function with ranking #

First, create a Postgres function that handles search and ranking:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
create or replace function search_books(search_query text)returns table(id int, title text, description text, rank real) as $$begin  return query  select    books.id,    books.title,    books.description,    ts_rank(to_tsvector('english', books.description), to_tsquery(search_query)) as rank  from books  where to_tsvector('english', books.description) @@ to_tsquery(search_query)  order by rank desc;end;$$ language plpgsql;

Now you can call this function from your client:

1
const { data, error } = await supabase.rpc('search_books', { search_query: 'big' })

Ranking with weighted columns #

Postgres allows you to assign different importance levels to different parts of your documents using weight labels. This is especially useful when you want matches in certain fields (like titles) to rank higher than matches in other fields (like descriptions).

Understanding weight labels#

Postgres uses four weight labels: A, B, C, and D, where:

• A = Highest importance (weight 1.0)
• B = High importance (weight 0.4)
• C = Medium importance (weight 0.2)
• D = Low importance (weight 0.1)

Creating weighted search columns#

First, create a weighted tsvector column that gives titles higher priority than descriptions:

1
2
3
4
5
6
7
8
9
10
-- Add a weighted fts columnalter table booksadd column fts_weighted tsvectorgenerated always as (  setweight(to_tsvector('english', title), 'A') ||  setweight(to_tsvector('english', description), 'B')) stored;-- Create index for the weighted columncreate index books_fts_weighted on books using gin (fts_weighted);

Now create a search function that uses this weighted column:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
create or replace function search_books_weighted(search_query text)returns table(id int, title text, description text, rank real) as $$begin  return query  select    books.id,    books.title,    books.description,    ts_rank(books.fts_weighted, to_tsquery(search_query)) as rank  from books  where books.fts_weighted @@ to_tsquery(search_query)  order by rank desc;end;$$ language plpgsql;

Custom weight arrays#

You can also specify custom weights by providing a weight array to ts_rank():

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
create or replace function search_books_custom_weights(search_query text)returns table(id int, title text, description text, rank real) as $$begin  return query  select    books.id,    books.title,    books.description,    ts_rank(      '{0.0, 0.2, 0.5, 1.0}'::real[], -- Custom weights {D, C, B, A}      books.fts_weighted,      to_tsquery(search_query)    ) as rank  from books  where books.fts_weighted @@ to_tsquery(search_query)  order by rank desc;end;$$ language plpgsql;

This example uses custom weights where:

• A-labeled terms (titles) have maximum weight (1.0)
• B-labeled terms (descriptions) have medium weight (0.5)
• C-labeled terms have low weight (0.2)
• D-labeled terms are ignored (0.0)

Using the weighted search#

1
2
3
4
5
6
7
// Search with standard weighted rankingconst { data, error } = await supabase.rpc('search_books_weighted', { search_query: 'Harry' })// Search with custom weightsconst { data: customData, error: customError } = await supabase.rpc('search_books_custom_weights', {  search_query: 'Harry',})

Practical example with results#

Say you search for "Harry". With weighted columns:

1. "Harry Potter and the Goblet of Fire" (title match) gets weight A = 1.0
2. Books mentioning "Harry" in description get weight B = 0.4

This ensures that books with "Harry" in the title ranks significantly higher than books that only mention "Harry" in the description, providing more relevant search results for users.

Using ranking with indexes #

When using the fts column you created earlier, ranking becomes more efficient. Create a function that uses the indexed column:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
create or replace function search_books_fts(search_query text)returns table(id int, title text, description text, rank real) as $$begin  return query  select    books.id,    books.title,    books.description,    ts_rank(books.fts, to_tsquery(search_query)) as rank  from books  where books.fts @@ to_tsquery(search_query)  order by rank desc;end;$$ language plpgsql;

1
const { data, error } = await supabase.rpc('search_books_fts', { search_query: 'little & big' })

Using web search syntax with ranking #

You can also create a function that combines websearch_to_tsquery() with ranking for user-friendly search:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
create or replace function websearch_books(search_text text)returns table(id int, title text, description text, rank real) as $$begin  return query  select    books.id,    books.title,    books.description,    ts_rank(books.fts, websearch_to_tsquery('english', search_text)) as rank  from books  where books.fts @@ websearch_to_tsquery('english', search_text)  order by rank desc;end;$$ language plpgsql;

1
2
3
4
// Support natural search syntaxconst { data, error } = await supabase.rpc('websearch_books', {  search_text: '"little puppy" or train -vegetables',})

Resources#

• Postgres: Text Search Functions and Operators