I’ve been playing with a demo over on MediaUnbound, which is a music discovery engine that runs in Flash and assembles pseudo-radio players that are designed to meet your every musical need. The concept isn’t particularly new, but I have to say I was impressed by the quality of the recommendations and – more importantly – by the process by which the recommendations were made.
Human beings have a tendency towards self-deception and care (to a greater or lesser extent) about the groups of people they are considered to be associated with. They care about how they are being compartmentalised or categorised by those around them and this – for the most part – is the way that most recommendations engines work: demographics, categorisation, association. In real life, people are reticent about revealing tastes that they’re slightly ashamed of and happy to own up to tastes that fit in with popular opinion or with the opinion of the subculture that they most heavily identify. Because of this, telling people what they might like is inherently a judgement by association – and if people aren’t comfortable with the way that the system has ‘judged’ them (either it seems clumsy and overly simple or it judges them to like music or books that are associated with cultures that they find tacky or crass) then they’re simply not going to be interested in hearing what it has to say. That’s as true online as it is when you recommend a song to a friend and they say that they don’t really like “that kind of music”, whether they’ve heard the song or not.
My experience with MediaUnbound has been better than with most other recommendations engines. Like the best LiveJournal quizzes the options you are provided with seem designed to make it a little more difficult to associate the choices you are given with ‘types of people’ you might or might not like. And the results give the impression of being satisfyingly varied between genres, cultures and styles allowing you to determine for yourself which recommended songs you think are central to your recommendations (ones you’re proud of or that you know you like or might find interesting) and which you can self-group as more peripheral (ones you’re ashamed of or don’t like). Add that to a generally pretty reasonable success rate and generally you can’t go that wrong… Here are a selection of some of the songs it thinks I’d like (I’m not going to tell you which ones are accurate or not – judge me as you see fit):
Sonic Youth: Silver Rocket
Porno for Pyros: Pets
David Bowie: Jean Genie
The Third Sex: At Least I Got Some Cool Clothes
7 Year Bitch: Dead Men Don’t Rape
The Pixies: Gouge Away
Spice Girls: Spice up Your Life
Eagle-Eye Cherry: Indecision
Weezer: Pink Triangle
Sleater-Kinney: I Wanna Be Your Joey Ramone
Cadallaca: Two Beers Later
The Breeders: Opened
L7: ‘Till the Wheels Fall Off
Hispana Tim: Sad and Lonely
Modest Mouse: Dramamine
Syrup USA: Vaporized
Tiffany: I Think We’re Alone Now
The Pixies: Silver
Tompot Blenny: Dr.Fitzpiers
Bruce Springsteen: Brilliant Disguise
The Breeders: Fortunately Gone
Confetti: Who’s Big And Clever Now
Razorcuts: Mary Day
Quasi: The Poisoned Well
Shop Assistants: All Of The Time
The Adverts: Bored Teenagers
Bedhead: Unpredictable Landlord
Britney Spears: Oops!…I Did It Again
Weezer: El Scorcho
7 replies on “On MediaUnbound and recommendations engines…”
MediaUnbound
Tom Coates alerted me to MediaUnbound. a new preferences engine for selecting music. I won’t comment much on the flash-based…
I can’t believe anyone can like both Spice Girls and Bruce Springsteen! I just can’t..
Mediaunbound’s good, innit? Coincidentally, I wrote a short note on music recommendations t’other day, in some way related to Clay’s ideas around situated software and social networks – I guess posing the question as to whether collab-filt-based recommendations are a better solution than building around small social networks … http://www.cityofsound.com/blog/2004/04/do_recommendati.html
It reminded me a bit of being at the opticians! Very clever, but it does seem slightly light on the dance music side.
I just discovered http://music.for-robots.com/ two days ago, and I ahd the same reaction you had to this. I just went through and downloaded all the recent MP3 files that have been posted, put them on my iPod and listened to them ñ†and I was surprised by how much I liked them. I had though I had my musical tastes pretty well defined (although there’s no simply word for it ñ I guess kind of Disco-ish Techno/Indie Rock/Folk Rock). But as I listened to these songs, many which were in categories I had more or less written off as being things I don’t like, I actually liked a lot of them.
> Human beings have a tendency towards self-deception
“That’s not true!”
the tiffany song got thrown in there, eh?
you need a way to do Amazon-like recommendations on these things. There needs to be a button that says “don’t figure this into the recommendation s/w”.