Re: Normalizing search scores over multiple indices

[Bill Janssen <ja...@parc.com>]

Hi, Otis.

Yes, I looked at Chuck's messages on the mailing list.  But he seems
to be focussing on MultiSearcher approaches, where I'm looking for a
way (I think) to adjust each individual index so that it will give
scores similar to the others.  I don't quite see how to merge the
results outside of Java without some such calibration.

Bill

> I think Chuck and friends have provided just such a patch, but we
> haven't applied it yet.... :(
> 
> Otis
> 
> --- Bill Janssen <ja...@parc.com> wrote:
> > I've got a situation where I'm searching over a number of different
> > repositories, each containing a different set of documents.  I'd like
> > to run searches over, say, 4 different indices, then combine the
> > results outside of Java to present to the user.  Is there any way of
> > normalizing search scores over multiple searches over different
> > indices?
> > 
> > Bill
> > 
> > ---------------------------------------------------------------------
> > To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
> > For additional commands, e-mail: java-user-help@lucene.apache.org
> > 
> > 
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
> For additional commands, e-mail: java-user-help@lucene.apache.org
> 



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

Re: Normalizing search scores over multiple indices

[Chuck Williams <ch...@allthingslocal.com>]

Bill Janssen writes (4/4/2005 6:38 PM):

>Hi, Otis.
>
>Yes, I looked at Chuck's messages on the mailing list.  But he seems
>to be focussing on MultiSearcher approaches, where I'm looking for a
>way (I think) to adjust each individual index so that it will give
>scores similar to the others.  I don't quite see how to merge the
>results outside of Java without some such calibration.
>  
>
There is also a long thread on this that pre-dates Wolf's solution for 
the MultiSearcher's.  I've never created the more general patch to make 
scores across separate indices normalizable.  In conversations with 
Doug, I concluded the best approach is to generalize the notion of a 
search score into a score object, thereby allowing multiple score facets 
(e.g., relevance as defined now, percent of query terms matched, etc.).  
The original idea was to use the percent of boost-weighted query terms 
matched of the highest result as the top normalized score (which fully 
defines the normalization).

Sorry it's not there...

Chuck


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