[By Subject] [By Date] [By Sender] [Prev] [Next]
Date: Wed, 3 May 1995 19:16:56 +0100
From: cockeril[at]-remove-icrf.icnet.uk (Matt Cockerill)
Subject: A survey of everyone's favorite bands?
>Now maybe will have some sort of answer to the question, "I like the jbc, what
>other bands might I like?"
Are you all aware that there is an experimental agent at MIT
(http://homr.www.media.mit.edu/projects/homr/), part of Patti Maes' work,
which is designed to recommend bands to you, once you have given a few
favourite/unfavourite bands marks out of ten. Works by statistical methods,
looking for your nearest members in terms of bands liked, and recommending
something which they like and which you haven't mentioned.
It sucks, by the way. Whatever the statistical methods they use, are, they
don't work, even if you input 100's of preferences. In fact, I think it
gets worse. The problem is that someone with eclectic tastes, which span
the available dataspace, gets reduced to a middle-of-the-road vector. i.e.
a combination of Bauhaus, Young Gods, Jazz Butcher, The Pogues and Babes in
Toyland averages out at Barry Manilow.
Matthew Cockerill <cockeril[at]-remove-europa.lif.icnet.uk> Tel: 071 269 3877
Imperial Cancer Research Fund Fax: 071 269 3801