Eric Wahlforss is one of the guys behind SoundCloud. This is where he jots down thoughts on the web, music and strategy, among other things.
More About Eric »
Towards the heavenly jukebox
A little over a month ago I visited last.fm in their very east Londonish office. Talking to the team, I got to know about some really cool things that they have in store for the future. Unfortunately I’m not allowed to talk about it here…
Mischa, Martin and Johan at last.fm.
One of the things they’re already doing though, is spidering (both manually and automatically) P2P networks, constantly feeding the radio database with terabytes of new material. Since they have a valid radio license, it’s fully legal!
This reminded me about what I wrote in 2003:
“Now, if file sharing apps would implement the cache model and rareness indexing I’m currently emulating manually, so that songs I never play are purged automatically from my local node and downloaded again upon request or automatically as their rareness indexes drops below critical, and if we could implement a way of auto-discovering other people with similar tastes by comparing my collection with theirs, it would make it even harder for commercial alternatives. Not to say if we’d come up with a cross-P2P search engine working on Google principles — “Yoodle”!”
Indy — a fairly new, client based “competitor” to last.fm — actually has a very simple caching functionality, resembling what I spoke of above. It basically dedicates more space for music relative to the rating you give it; 5 star music gets 1GB, 4 star gets 500MB, 3 star gets 100MB, etc. The cache is then FIFO-purged automatically.
Alas, I never got convinced by Indy since the music it plays simply is too bad. I believe the base of CC-licensed music actually isn’t there yet in terms of quality. I ended up pressing the 1 star button for every track, desperately hoping for the collaborative filtering system to find something good for me — but it never happened. I think I had one 3 star track out of 25 bad ones…
So I’m back to last.fm for now, thinking it would be very cool if we could somehow integrate last.fm with Indy, getting the best of both worlds.
Structured Blogging and Collaborative Filtering
In 2003 Alf Eaton proposed a standard RDF-vocabulary for reviews. That started off an interesting discourse that has now lead up to the Structured Blogging Initiative (among other things). Structured blogging is basically about..well..structuring blog content (ratings, reviews, events, etc) so that it becomes machine-readable.
It struck me that one of the cool things that one could potentially do — if a lot of people start to use this technology — is to build a distributed collaborative filtering system on top of it. I imagine a service that would track your blog, get out your ratings/opinions, build a profile for you (much like last.fm does), and match that with the profiles of other bloggers. The system could then provide you with a number of feeds that could give you relevant and accurate recommendations on things such as books, movies and music. It could also point you to other bloggers with similar taste.
This technology — again if it gets widely adopted — will, I believe, have interesting effects on the blogosphere. What happens when there is a massive peer-to-peer colloborative filtering infrastructure around? One thing is certain — The tail will get ridiculously long…
UPDATE: This 2002 paper by Hassan Masum outlines TOOL, The Open Opinion Layer. Great read.
Linkratings
Jyri has cool UI ideas for newsreaders. The question I have is: who should take care of the rating input in that pretty little box…?
What I’d like to see is a simple service that could store my ratings for links. The service would collect rating data from many users and then apply collaborative filtering techniques, providing me with feeds of links I’m guaranteed to like…