Archive of UserLand's first discussion group, started October 5, 1998.

Re: Collaborative Filter theory.

Author:- -
Posted:11/10/1999; 11:39:16 AM
Topic:Collab filtering at my.userland.com?
Msg #:12938 (In response to 12914)
Prev/Next:12937 / 12939

But we're not talking about search engines, we're talking about collaborative filters. Large quantities of data are more important for a collaborative filter than the cleverness of the algorithm.

And attempting to implement *any* kind of collaborative filter will have you sitting crosswize of somebody's patent: the CF patent situation is a garbled mess; the US Patent Office has issued patents for what is basically handwaving and statements of CF theory.

There's a remarkable amount of guessing going on in general. Allow *me* to cite a specific example: www.alexlit.com. After providing specific opionions on a select number of works, you're presented with (by default) twenty titles of works of literature you're predicted to enjoy, ranked by a combination of the level of enjoyment and confidence in the prediction. If you happen to want more than twenty, you can click "next" and get the next twenty out of a list that typically includes thousands of titles.

Sorting *is* screening. There's no theoretical benefit to "protecting" somebody from the less close matches: assume they're intelligent enough to figure out for themselves when the results have crossed the line of irrelevance. They can do a lot better job than you can in most cases.

www.alexlit.com, BTW, is my site. From extensive testing, I can quite categorically state that more data is more important than clever refinements to the algorithm.


There are responses to this message:


This page was archived on 6/13/2001; 4:53:27 PM.

© Copyright 1998-2001 UserLand Software, Inc.