Mailing-List: contact java-user-help@lucene.apache.org; run by ezmlm
Precedence: bulk
Reply-To: java-user@lucene.apache.org
Received-SPF: pass (nike.apache.org: domain of MSchraeder@btsb.com designates
 216.8.91.5 as permitted sender)
Message-Id: <s9dcab94.070@mail.btsb.com>
Date: Wed, 08 Apr 2009 13:50:00 -0500
From: "Matt Schraeder" <MSchraeder@btsb.com>
To: <java-user@lucene.apache.org>
Subject: Re: Suggestive Search
Mime-Version: 1.0
Content-Type: multipart/alternative; boundary="=__PartC4EC5CC8.0__="

--=__PartC4EC5CC8.0__=
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit

Corerct me if I'm wrong, but I don't think n-grams is really what I'm
looking for here.  I'm not looking for a spellchecker or phrase checker
style suggestive search, but only based on the exact phrases the user is
currently typing.  Since Lucene uses term-based searching, I'm not sure
how to have it search on portions of a full phrase.  Using a standard
lucene search typing in "harr" will result in searching for "harr" as a
term, which will not find "Harry Potter".  Using ngrams it would find
"Harry" as a term, but not at the beginning of an entire phrase.  This
would bring back "My Dog Harry" as a result, which isn't what I'm
looking for. I just want phrases from fields beginning with "Harr"
only.
 
I could easily do this all with our database server by simply doing a
query for "where searchqueries like 'harr%'" but we're trying to limit
our hits to the database to keep speed up on the site.

>>> karl.wettin@gmail.com 4/8/2009 12:49:45 PM >>>

For this you probably want to use ngrams. Wether or not this is  
something that fits in your current index is hard to say. My guess is 

that you want to create a new index with one document per unique  
phrase. You might also want to try to load this index in an  
InstantiatedIndex, that could speed things up quite a bit if the  
corpus is not too large.

If your suggestion text corpus is really large and you only want  
forward-only suggestions then you might want to consider a trie- 
pattern solution instead. These can be rather resource efficient, even 

when loaded to memory.

If you have a lot of user load on your search eninge then it might be 

interesting to use old user queries as the base of your suggestions  
and perhaps boost a bit on trends, i.e. the more people search for  
something the more it get boosted in the suggestions list.


      karl

8 apr 2009 kl. 15.26 skrev Matt Schraeder:

> I want to add a suggestive search similar to google's to
autocomplete
> search phrases as the user types.  It doesn't have to be very  
> elaborate
> and for the most part will just involve searching single fields. 
How
> can I perform a search  to be able to fill in autocomplete text?
>
> For instance, if I start typing "Harr" it should bring up "Harry
> Potter" "Harry Houdini" and "Harry S. Truman"
>
> I have tried doing search queries for "Harr*" but it's still doing
> term-based searching rather than searching a full field.  To make a
> field both searchable as the full field as well as tokenized, would
I
> have to duplicate the field and make one a keyword field? Is there a
> more convenient way to do this? I have also considered making a
second
> index for suggestive search, which would only have the fields that I
> want to enable suggestive search on, but this seems like it would be
> unneccesary duplication of data as well, though it would probably
make
> suggestive search faster due to a smaller index.
>
> Ideally it would also be nice to be able to rank these terms based
on
> the number of times they have been searched for so that the results 

> are
> tailored more to our users rather than simply just the score that  
> Lucene
> chooses.


---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-user-help@lucene.apache.org


--=__PartC4EC5CC8.0__=--