Sorry, I used AND-statements instead of OR-statement in the example.
I notices that gin is much faster than gist, but I don't know why.
The query gets slow, because there are many non-stop words which appear very often in my sentences, like in 3% of all the sentences.
Do you think it could be worth it to filter the words, which appears that often and declare them as stop-words.
How would you split a sentence with let's say 10 non stop words to provide a performed similarity search?
There's still the problem with very short sentences. An partiel index on them with the trigram search might be the solution.
The pg_trgm module is far to slow for bigger setences, like you showed.
I thought I'll build a few partiel indexes on the string length, to enhance the performance.
Do you know some more improvements?
Janek Sendrowki