I was going to avoid talking, because I can't remember the name for this
exactly, but its a pretty basic AI principle. I had a good time talking
about it with a nice guy over the lunch break from the IJCAI conference that
was in Seattle not too long ago.
Basically, you take a set of one users's choices on something. You then
check for other people that have rated several of the same items, and give
them an index based on whether they agree with that user or not on the rated
items they share. Then, you might want to normalize that users choices, as
some people just like everything :) Next, you pick some of the most similar
ones, grow their result to items that our target user hasn't seen yet, and
return those results in order of (user_index * user_rating). Viola!
I can't believe we give people patents on these things. Next think you know
they'll have the left human leg patented, and will be wanting money every
time I take a step...
Kevin Watt, AllPoetry.com Community manager: Poets Unite!
"Here, write it, or it will be erased by the wind." - Isabel Allende
From: Jesus Quiroga
Sent: Tuesday, August 28, 2001 3:34 PM
To: Terrence Monroe Brannon
Subject: Re: Perl Personalization Engines
At 18:57 28/08/2001, you wrote:
I've been sitting here trying to checkout my purchases at amazon.com for
the last 30 minutes now and every time I'm about to log out, some new
Perl/Apache book catches my eye in the sidebar and I end up adding it to
my cart too... I hope to be checked out by noon at this rate. :)
Amazon.com uses Perl heavily, but the exact technologies behind their
unrivaled personalization engine are of course not disclosed.
Actually they disclosed something. You will be interested in these US
patents granted to Amazon.
-US6266649: Collaborative recommendations using item-to-item similarity
-US6064980: System and methods for collaborative recommendationshttp://www.delphion.com/details?pn=US06064980__