Skip to main content.
home | support | download

Back to List Archive

Re: Title ranking

From: Bill Moseley <moseley(at)not-real.hank.org>
Date: Thu Dec 27 2001 - 16:56:00 GMT
At 07:43 AM 12/27/01 -0800, Bob Stewart wrote:
>Bill,
>
>I've been using Atomz.com's commercial remote search solution. They have
one feature that might be worth emulating. They have a particular meta tag
that trumps even the title. So a document with a word listed in this meta
tag will rank even higher than one that's listed in the title. I guess it's
the same as the meta keyword tags that used to work with some of the portal
search engines.

Currently, you can adjust the rank for meta tags in config.h, but not per
metaname.  We have discussed this specifically.  It would be something like

MetaNameRank  foo 4
MetaNameRank  bar 3

It's really complicated, though.  Rank should depend on a number of things:

  o number of words in the document that get hit
  o what meta tag it's found in
  o what HTML it's in <h1> might rank higher than <b>
  o word proximity
  o how close the word is to the top of the document???
  o size of document?
  o probably also something I forgot

So, anyone know how to put that into an equation?

A note on word proximity -- looking at search logs shows that often
multi-word searches are often times really phrase searches without quotes.
So, when searching more than one word, docs with the words close together
should rank higher.  But, there's a cool side effect of ranking by word
proximity:  Swish could use OR as the default search, but for multi-word
searches and proximity ranking that pushes all the ANDs to the top of the
rank.  This means the best of both the AND and OR "worlds".  And since
swish is often used to index small sets of files, OR search would be good
as that would result in fewer "no results".

Funny you should mention Atomz.  I was looking at it the site just a few
days ago (http://www.atomz.com/search/).  I like the idea of an off-site
search service.  Might as well out-source if you have the money and don't
have the ability to do it in-house.  Swish-e is on the other end of the
spectrum there.

I also thought that reading their features on their site that it sounded a
lot like swish.  I've often thought about that type of site format for
swish's site -- makes it reasonably easy to get an overview of the "product".

But, I did kind of chuckle at their site, too.  Like those ROI and TCO
pages.  I guess IT managers need numbers and charts.  Isn't their search
service something like $10,000 a year?  And they still get 404% (not 400%
mind you) ROI.   Not too bad "ROI", I guess.  (How that applies to their
search software is not clear).



-- 
Bill Moseley
mailto:moseley@hank.org
Received on Thu Dec 27 16:56:07 2001