Skip to main content.
home | support | download

Back to List Archive

Re: Swish-e Ranking for Search within XML files

From: Peter Karman <karman(at)not-real.cray.com>
Date: Mon Sep 13 2004 - 14:15:48 GMT
If you have the latest cvs build, try using the -R 1 option (or 
RankScheme( 1 ) if using SWISH::API) and see if that makes a difference.

You might try searching for 'test AND code=test' to see if that helps 
make your results smaller (though that may not be what you want).

Tomek NoiƄski wrote on 9/12/04 1:45 PM:

> On Fri, 10 Sep 2004 13:28:02 -0700 (PDT)
> Peter Karman <karman@cray.com> wrote:
> 
> 
>>>It does work, but I wonder if there is a way to make meta name
>>>behave as it would have the rank stronger than 10?
>>
>>because it is an experimental feature, I suggest you experiment and
>>let us know. :)
>>
>>you can redefine the range here:
>>
>>swish.h:#define RANK_BIAS_RANGE 10
>>
>>set it to something larger than 10 and recompile swish-e. Then let us 
>>know if a higher range is an effective tool.
>>
>>the most recent CVS daily build has some other ranking improvements,
>>so you might experiment with that version for this test, to see if
>>those changes help with your biasing need.
> 
> 
> Thanks, I tried it.
> Changing to newer version (Debian 2.4.1-1 -> today's snapshot) helped,
> but I didn't have luck with MetaNamesRank with changed RANK_BIAS_RANGE.
> I didn't seem to matter if MetaNamesRank was set to 10, 100 or 1000.
> 
> Maybe there's better way to do what I'm trying to do? It's not too
> important, but I'm curiout how to do it right.
> 
> I have a base with algorythmic problems, with descriptions and some
> other data, like "code", identyfying the problem, author etc.
> I form it into xml, make an index from it and do queries like
> "test or code=test".
> The problem is, problem with code "test" has much lower rank than
> another, mentioning several "test cases" in the description, despite
> MetaNamesRank set to 10/100/1000 for code (but it does work somewhat,
> because when set to 0, it is even worse).
> 
> Should I try to sort the records myself and not rely on swishes ranking?
> Or just wait for the MetaNamesRank to become stable? :)
> 
> Oh, and thanks for help and a great piece of software!
> 

-- 
Peter Karman  651-605-9009  karman@cray.com
Received on Mon Sep 13 07:16:33 2004