[Xapian-discuss] Modifying Result Relevance

Olly Betts olly at survex.com
Tue Mar 29 15:31:20 BST 2005


On Tue, Mar 29, 2005 at 09:17:57AM -0500, Mike Boone wrote:
> My client has decided that he wants to give the highest relevance to
> documents that contain the highest number of search terms, with no regard
> for the document length or other factors. I don't really like that approach
> but I do my best to keep them happy. Is there any way to accomplish this
> search result scenario with Xapian?

This is what you'll get from BM25 with large k1 and k2=0, k3=0, b=0.
In the TREC tests, "large k1" is usually 10000.

> Perhaps with the weighting stuff, but I seem to recall that it wasn't
> accessible from the PHP interface...is that correct?

You can't derive your own subclass of Xapian::Weight in PHP, but I
believe you can create a BM25Weight object and tell Enquire to use it.
If not, it shouldn't be hard to make that work.

> (P.S. I've been reluctant to upgrade Xapian with all the PHP/SWIG discussion
> going on...is it safe yet?)

That's all in CVS HEAD.  The released versions should work (but use
xapian-bindings 0.8.5.1 rather than 0.8.5).

Cheers,
    Olly



More information about the Xapian-discuss mailing list