lucene-java-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From karl wettin <ka...@snigel.net>
Subject Re: Lucene and learning search
Date Wed, 07 Jun 2006 10:25:09 GMT
On Tue, 2006-06-06 at 22:23 +0000, michael turner wrote:
> Working on a project that requires a Search query similiar to what is 
> seen on"amazon.com" in that after searching for and displaying an item, 
> the system shows:
> 
> "Users that have searched for "AAAAA" AND "BBBBB" have also searched 
> for "CCCC".
> 
> Where "BBBBB"  and   "CCCC" are other related entities.
> 
> 
> 1. Has anyone used or heard of this algorithm? 

A quick and dirty reply:

Actually, Amazon does not track the queries, they track what customers
buy. The "algorithm" you are looking for is collaborative filtering. 

You would probably want to build a classification scheme much more
detailed than the formula 1-a facets in order to get good results.

To find such clusters based on searches is tricky as you only want to
take the results that was relevant in to account.

If you only want to find some statistics of how people refine or change
their queries to find what they are looking for, then you need to know
the original query and the end query, and be really sure that the
subject of the search did not change. Overlap all sessions you saved and
use some statistical analysis library (perhaps Weka can help you?) to
get the results.

It might also be possible to implement using hidden markov models.

You really have to tell us more about the application you add Lucene to
in order to get a good reply on how to implement a suggestion scheme.


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


Mime
View raw message