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

Re: Collab filtering at alexlit.com

Author:Paul Snively
Posted:11/7/1999; 4:57:31 PM
Topic:Collab filtering at my.userland.com?
Msg #:12798 (In response to 12793)
Prev/Next:12797 / 12799

Anita Rowland wrote:

Alexandria Digital Literature http://www.alexlit.com/ uses collab filtering.

As you rate things that you've already read, it can make better and better recommendations about things you might also like.

I think collaborative filtering technology that requires people to "rate" things is doomed to fail (e.g. Firefly).

What Amazon does works: it correlates what people buy and makes recommendations based on that. End of story. Nothing fancy about it.

How this might work in the absence of an explicit commissive action such as a purchase decision might involve, e.g. weighting the correlations based on amount of time spent viewing pages (and assuming, of course, that you have some fairly good categorization mechanism for what each page is "about." We're back to metadata again, as always: some time ago, in this very forum, Jakob Nielsen pointed to the LSA research being done at the University of Colorado as a possible domain in which to address automatic generation of metadata for English text). Similar weighting could/should be done based on how many people who spent amount of time X on page Y also spent an amount of time within Z% of the amount of time user U spent on page P that's been correlated with page Y.

The system is likely to start out badly overgeneralizing, but as more categories are identified (either by hand or by tools like LSA) and more people use the system, the weighting should--in theory--make the correlations more and more accurate over time. No questionnaires needed.

There are responses to this message:

This page was archived on 6/13/2001; 4:53:23 PM.

© Copyright 1998-2001 UserLand Software, Inc.