rank algoritme - best bet search

From: <jetpoet_at_yahoo.com>
Date: 6 Jun 2003 12:49:06 -0700
Message-ID: <9d093ada.0306061149.7263866c@posting.google.com>

Hi All!

I would like to know if any of you know of a way to create a ranking algorithm for a database search.

I want to allow my users to give an element a rank from -5 to +5 and based on that the system should make recommendations of other elements that might be of interest. This recommendation should be based on your own and other users rankings. I am thinking of two scenarios that I would like to implement, but either one will do if only one is possible.

When browsing elements, the site should be able to say: "people who liked this element also liked these elements". Kinda what Amazon does. This is only based on what people liked the particular element. This should of course be weighted, so people who ranked this element high carry more weight in the ranking of other items.
"Based on what you have said you like we recommend this." This is what I think is harder to implement. People who have said they like the same things as I do are weighted higher in their ranking recommendations, but if there are conflicts (ie. people who rank something high that you have ranked low) their weight should be decreased.

The table I have imagined should hold this ranking information is quite simple.

RANKING

USER_ID
ELEMENT_ID
RANK But calculating the recommendations is anything but simple. I have thought of writing some stored procedures that extract lists of weighted USER_IDs that have ranked an element, and then using those as the basis for my calculations, but it turns out really nasty.

Does anyone have any idea of where I should start. Can anyone tell me of an algorithm that solves this problem.

I can say that I at the moment have 4,000 users who on a regular basis read over 100,000 posts. I want them to quicker be able to find what they are looking for, and be able to find relevant information.

Thankyou in advance
JP Received on Fri Jun 06 2003 - 14:49:06 CDT