Skip to main content.
home | support | download

Back to List Archive

SWISH-E Ideas

From: Markus Peter <warp(at)not-real.spin.de>
Date: Mon May 10 2004 - 16:49:04 GMT
Hello

We're using SWISH-E as search engine for a couple of projects. Here are two 
ideas which came to us the last couple of months.

1) Modifying properties
It would be nice to be able to modify properties in an index without 
re-indexing even with SWISH-E running non-incrementally. This should not be 
too much of a problem to implement if the type of such properties would be 
limited to fixed-width properties, e.g. a simple integer, and those 
properties would be stored unsorted which SWISH already supports.

This helps for a lot of situations where one would normally desire to use a 
fully incremental system, but where it's actually not necessary. Example: 
We use SWISH-E to index mails. For the user, it's not that big a problem, 
if newer mails are not included in the index yet, but it's annoying if he 
uses the search system to actually delete mails which then still show up in 
the search results. Modifyable properties would allow us to just set a 
deleted flag in the index for such a mail and -L could be used to skip them 
in the results.

We could of course store such flags externally, but then things quickly 
become complicated if we page the result sets.

2) Modifying the ranking dynamically
It would be nice if a property could be assigned via the config file which 
influences the ranking of a document. This could also be a simple floating 
point number property, where the property is multiplied with the ranking 
SWISH-E calculates for a search. This allows for all kinds of flexible 
modifications of ranking, like e.g. "PDFs are usually more relevant than 
HTML pages", or building indexers which take user ratings, link counts etc. 
into account. I can of course at the moment already store such a value in 
the database and use it for sorting, but then the value would have 
precendence over the ranking, which is not what I'd like to achieve 
(modifying a hit's ranking +/- 20% but still relying mainly on SWISH's 
ranking), or I could apply such calculations afterwards, but then I again 
run into trouble with paging.

-- 
Markus Peter - SPiN AG
warp@spin.de
Received on Mon May 10 09:49:04 2004