使用smlar插件, 相似度计算公式采用tfidf， 效率和你列出的差别特别大 about blog HOT 2 OPEN

我参考的是你这篇文章的方法：
https://github.com/digoal/blog/blob/c1a3db101af25ababd35d0e6963e4c1c4609c8d5/201701/20170116_03.md

from blog.

set enable_bitmap_indexscan=false;
只走index scan，查询计划

                                                              QUERY PLAN                         

Limit (cost=416.94..416.97 rows=10 width=258) (actual time=3561.033..3561.033 rows=0 loops=1)
Output: document_id, body, (smlar(regexp_split_to_array(lower(body), '[^[:alnum:]]'::text), '{fringilla,mal
esuada,euismod}'::text[]))
Buffers: shared hit=13022
-> Sort (cost=416.94..417.19 rows=100 width=258) (actual time=3561.032..3561.032 rows=0 loops=1)
Output: document_id, body, (smlar(regexp_split_to_array(lower(body), '[^[:alnum:]]'::text), '{fringil
la,malesuada,euismod}'::text[]))
Sort Key: (smlar(regexp_split_to_array(lower(documents.body), '[^[:alnum:]]'::text), '{fringilla,male
suada,euismod}'::text[])) DESC
Sort Method: quicksort Memory: 25kB
Buffers: shared hit=13022
-> Index Scan using documents_tokenize_idx on public.documents (cost=0.28..414.78 rows=100 width=25
8) (actual time=3561.019..3561.019 rows=0 loops=1)
Output: document_id, body, smlar(regexp_split_to_array(lower(body), '[^[:alnum:]]'::text), '{fr
ingilla,malesuada,euismod}'::text[])
Index Cond: (regexp_split_to_array(lower(documents.body), '[^[:alnum:]]'::text) %% '{fringilla,
malesuada,euismod}'::text[])
Rows Removed by Index Recheck: 12299
Buffers: shared hit=13022
Planning time: 0.247 ms
Execution time: 3561.105 ms

from blog.