Edge NGram search in PostgreSQL

Question

I need to make search-as-you-type autocomplete for a large list of companies (over 80,000,000). The company name should contain the word that starts with a search query like this

+-------+----------------------------------+
| term  | results                          |
+-------+----------------------------------+ 
| gen   | general motors; general electric |
| geno  | genoptix; genomic health         |
| genom | genoma group; genomic health     |
+-------+----------------------------------+

The pg_trgm module and GIN index implement similar behavior but don't solve my problem.

For example, ElasticSearch has feature Edge NGram Tokenizer that completely fits my requirements.

From documentation:

The edge_ngram tokenizer first breaks the text down into words 
whenever it encounters one of a list of specified characters, 
then it emits N-grams of each word 
where the start of the N-gram is anchored to the beginning of the word.

Edge N-Grams are useful for search-as-you-type queries.

Is there a similar solution in PostgreSQL?

Wingate

Answer 1

I create a custom tokenizer

CREATE OR REPLACE FUNCTION edge_gram_tsvector(text text) RETURNS tsvector AS
$BODY$
BEGIN
    RETURN (select array_to_tsvector((select array_agg(distinct substring(lexeme for len)) from unnest(to_tsvector(text)), generate_series(1,length(lexeme)) len)));
END;
$BODY$
IMMUTABLE
language plpgsql;

This function creates all edge ngrams like this

postgres=# select edge_gram_tsvector('general electric');
                               edge_gram_tsvector
-----------------------------------------------------------------------------------------
 'e' 'el' 'ele' 'elec' 'elect' 'electr' 'g' 'ge' 'gen' 'gene' 'gener' 'genera' 'general'
(1 row)

Then I create a GIN index for tsquery

create index on company using gin(edge_gram_tsvector(name));

The search query will look like this

b2bdb_master=# select name from company where edge_gram_tsvector(name) @@ 'electric'::tsquery limit 3;
                    name
--------------------------------------------
 General electric
 Electriciantalk
 Galesburg Electric Industrial Supply
(3 rows)

Performance of solution is quite high

---

explain analyse select * from company where edge_gram_tsvector(name) @@ 'electric'::tsquery;


Bitmap Heap Scan on company  (cost=175.13..27450.31 rows=20752 width=2247) (actual time=0.224..1.019 rows=343 loops=1)
  Recheck Cond: (edge_gram_tsvector((name)::text) @@ '''electric'''::tsquery)
  Heap Blocks: exact=342
  ->  Bitmap Index Scan on company_edge_gram_tsvector_idx  (cost=0.00..169.94 rows=20752 width=0) (actual time=0.138..0.138 rows=343 loops=1)
        Index Cond: (edge_gram_tsvector((name)::text) @@ '''electric'''::tsquery)
Planning Time: 0.216 ms
Execution Time: 1.100 ms

Wingate

This is quite good. Were you able to add relevance ranking? I'm trying to figure out how to use ts_rank_cd with this scheme. – Tolu