One interesting thing that I found with building the recommendation engine is that I got much more pleasing results by considering each person's tag independently
rather than the person as a whole.

However, from a UI standpoint, I'd like to provide people with a way to get a global view of who is recommended to them in general, presumably by aggregating the per-tag
recommendations (currently too slow to be done.)

Another issue I've been thinking about is the notion of influence in general; how to quantify the metric in some reasonable way? For example, if an item was posted 10 times, and person X was first, was he influential? What if the other 9 times were the same day? Or two weeks laters? What
if 1000 more people were to go on to do it?

I suppose something interesting might occur if you just looked at a user's urls and score every other user who bookmarked that URL before them. Alternately a bayesian predictor to suggest if person X posts
something, person Y is likely to.

Shame about all the I/O this would take up. Probably time to order more servers anyway.

Joshua

On Dec 5, 2005, at 11:25 AM, Chris Lott wrote:

On 12/5/05, Brian Del Vecchio <[EMAIL PROTECTED]> wrote:
I'd rather have
tools that help me sift through the river for valuable items.  I find
the recommendations very useful, and would love to see more work like
this.

I agree. I'm not interested in "who is the most influential" though I
am VERY interested in "who is the most influential when related to
Chris' links." The former is a game, the latter is a tool for finding
new information. One of the reasons I have stayed with del.icio.us is
that I have a fair amount of trust that Joshua wants to focus on the
usefulness and leverage of the social network effects to uncover and
recover information.

c
_______________________________________________
discuss mailing list
[email protected]
http://lists.del.icio.us/cgi-bin/mailman/listinfo/discuss


--
joshua schachter
[EMAIL PROTECTED]


_______________________________________________
discuss mailing list
[email protected]
http://lists.del.icio.us/cgi-bin/mailman/listinfo/discuss

Reply via email to