I used to spend a lot of time explaining to clients how powerful website personalisation was going to be for them and researching the technologies to do it. I remember one day, the people at Broadvision (a personalision software company) invited me to a seminar on the topic and showered me and a bunch of other agency folk with quality baked goods and average coffee.
Anyways, I was reminded of this last week when someone at work said something that sounded exactly like an old Broadvision pitch and we all laughed and got to thinking about what happened to all those amazing self-adjusting sites we going to build. In the end, they just didn't happen.
That was this Friday, and I just got my newsletter from OK/Cancel today. Tom Chi asks
There was a period in the history of the web when personalization was going to transform everything -- making the buying experience vastly more inviting for customers and lucrative for business. I'm not really sure what happened.
He's spot on as ever. Read Getting Personal
Comments (1)
"Chris":http://chris.carline.org/ had this to add:
Hi Dug
Noticed your broadvision/personalisation post, and then this popped up on one of my mailing lists. It's only vaguely related, but I thought you might be interested to read about collaborative filtering mechanisms (such that Amazon use to suggest new stuff)...
Chris
---------- Forwarded message ----------
Subject: Re: (void) Internet radio
On Thu, Jan 20, 2005 at 11:05:22AM +0000, alex said:
_I think they tweak the algorithm every now and then but the problem seems to be that collaborative filtering gives you a dull average, and other people always like crap music. It does turn up nice surprises and introduce me to new music but not nearly as often as I would expect._
I think I've ranted about this before but, apparently, this is a problem with Collaborative Filtering in general.
A fe wyears ago I knocked up a really simple system, seeded from IMDB, that recommended films.
The problem was that it worked too well - it recommended films that i already liked. Which meant that it was working but useless.
So I tried throwing in some randomness. Which made the system fail completely and start throwing back junk.
So I joined a Collaborative Filtering mailing list and asked what was going on and got an off-list reply from someone who explained that this happens to all totally automated CFs. There is a suspicion that it's an Unsolveable problem because, in essence, what you're tryign to do is achieve two different goals - map a trend and also deviate it from it. Or something. There was maths, I glazed over. Somebody may have mentioned systems with Chaotic tendencies. Or neutral aligned good or something.
I've been on the list since then and apart from an occasional conference CFP it's pretty quiet. And about every 7 months somebody turns up and says "My CF is *almost* working but ...."
Posted by Bozo
|
January 21, 2005 10:11 AM
Posted on January 21, 2005 10:11