Felix Miller talked about last.fm at ETech. What follows is my as-it-happened transcription, which hasn’t been proofed or edited. My apologies to Felix for any inaccuracies.
Felix Miller from last.fm at ETech, 7 March 2006
Thanks for being here. I am one of the cofounders of last.fm, here to talk about The Musical Myware. I think I’ll be the first person here to talk about the attention economy, it’s all about how to use attention data to create useful products and services.
What is attention data? It’s about stuff that people pay attention to and don’t pay attention to. We just have to record this by observing what people do. We can now do myware, which sounds like spyware but isn’t. It’s the useful act of someone spying on themselves.
Why should people be spying on themselves? Why should they share their data with a company like last.fm? What do I get out of this? We want people to tell us what music they’re listening to, at any time, all the time, without realizing it. Why are we doing it? Why are our users doing it?
We’re doing it because Napster showed us that all music that was ever recorded was available for me to listen to and play around with. People will need to know what they should be listening to. Our mission is harnessing the knowledge of the crowd. Everything that you need to know about your music consumption is already known by someone else out there.
Developed AudioScrobbler that installs into your PC’s music player, reporting listening metadata to last.fm. Only records what you listen to, doesn’t record the music you skip. If you didn’t listen to a track all the way to the end, it isn’t recorded.
An event (“I listened to this song”) is a submission. We get 8M submissions per day. Over a billion in the last year, and who knows where we go from here. Only by installing this software into people’s applications can we get knowledge data.
All this track information goes into people’s music profiles, a massive listening history of what people are listening to. While they’re listening to music, they’re building our track catalog. When you submit a song that the system doesn’t recognize, it creates a new entry in the track. 8M tracks in database, total size of track catalog would be 25M. When it comes to music, we think we have the longest long tail.
What do we do with all this data? How do we datamine this knowledge of the crowd? We use collaborative filtering to generate the base layer for our recommendations. User-to-user similarity (neighbours) and artist-to-artist similarity. From this, people can start browsing, finding new music, jump from artists to profiles to songs and so on.
We have popularity data for every item in our catalogue. For a band like Los Coronas, we know exactly which of their tracks are most popular and what people are listening to.
Every person gets their own profile page, with listening charts for overall and last week. Can point you to people with similar tastes to you. By going to their profiles, you can find similar music.
System creates personal music recommandations every week, based on what you listened to last week. Because we have popularity data, can filter the recommendations from most popular to most obscure.
Music taste is complex. Introduced tagging to help people make sense of their profile and help us make sense of the data. Can tag artist, album, or track. Shows your tags, tags most popular for this item. Getting people to reuse tags and getting people to describe things in the same way. Helps us organize our universe and improve recommendations.
We stream radio where everyone’s music tastes is a radio station. Can go to artist radio similarly (all things like this artist shuffled into a radio stream), or by tag, or by group. People are tagging their own profiles and people are creating their own radio stations.
If you listen to something you like it, skip it you don’t like it, love it button = i love this music, ban = i hate this, never again please. We stream in 128kbps MP3, and all software we produce is open source and attempt to support all the OSes. We want to level the playing field for musicians and artist to promote their stuff. Shouldn’t need a major record label contract or massive promotion budget to fidn the right audience for your music.
Getting back to question I started with: why should people spy on theirselves? Provide a valuable and compelling service and so users will shjare their attention data with you. Users must be in control. We supply XML feeds from profiles, so you can bring your data into your blog. All CC-licensed. Anything you don’t like, you can edit out, and you can destroy your data on our system whenever you want.
Time to put the listener in charge. That’s what we call the social music revolution. Thank you very much.