Обсуждение: Searching a DB index.. possible?

On 05/02/10 15:48, Moe wrote:
> I am trying to figure out how I can introduce full text search "tips"...
> tips meaning, like if you start typing Mer ..
> then a box with a couple of words underneath comes up, with *Mercedes,
> Mercury *and so on :P
>
> I am thinking that the word suggestions needs to come from the already
> existing tsearch2 index.
> Words that are frequent in the index could be ranked higher as a search tip.

This could get expensive if you have a large index. You might want to
summarize down to a separate (word,frequency) table. Particularly if you
don't want stemming to interfere with your suggestions.

See the manuals (full text search) for details on prefix searches -
introduced in 8.4 if you really want to go the tsearch route.

--
   Richard Huxton
   Archonet Ltd

On 11/02/10 17:11, Moe wrote:
>> You might want to summarize down to a separate (word,frequency) table.
>> Particularly if you don't want stemming to interfere with your suggestions.
>
> Stemming is ok. That's fine.
>
> How would this normally be implemented?  Should I go through my text word by
> word myself, or the already existing index and copy over to a new table? Not
> sure how this would be managable. Every change to something would require a
> word count down, and then a word up after.

I wouldn't bother keeping it 100% up-to-date. You only want the
most-plausible suggestions anyway. Have a look at the ts_stat() function.

http://www.postgresql.org/docs/8.4/interactive/textsearch-features.html

Run it overnight, and summarise to a separate table, then just use
"like" against it.

Otherwise, like you say, your database will spend most of its time
adding up word counts etc.

--
   Richard Huxton
   Archonet Ltd