APML isn't just for humans*
We have recently been looking at (Attention Profiling Mark-up Language). APML allows you to share your own attention data. That's data about what you have given your attention to; whether by browsing websites, reading RSS feeds or listening to music. You could then take your attention data profile and pass it to another website which would then be able to automatically customise itself to your preferences or interests. That seems to be its conventional use-case.
Michael from our team has been looking at this from an alternative direction. What if you could generate APML for a music radio show? That would be based on the music that the show or DJ has played and, by extension, has been paying attention to. So Michael has hacked up some APML feeds for some of our radio shows based on their tracklistings (warning - these are very beta and we cannot guarantee they are accurate or stable).
And he's also generated APML for John Peel, based on the artists that played in sessions on his show from 1967 to 2004. Michael is a big fan of The Fall and it's not necessarily a coincidence that they come out top...
We tried feeding these into , a personalised music magazine that accepts APML, and after some teething troubles this now works. And we've also put them into Sun's , an experimental site for generating and processing attention data, and they've kindly generated some recommendations based on the radio show profiles.
A couple of things that this has thrown up around APML. One, there is no validator for APML out there, if you know of one then let us know as it would be most helpful. And we also think that the concepts in APML files should allow URIs to identify them more accurately. At the moment they just have a text attribute describing the concept but the addition of a URI that could point to a resource on or wikipedia would help better define the concept involved.
What else could we do? Well, if we tracked what radio programmes people listened to then we could generate APML for them based on what they've listened to. This could be both for music and for speech once we've got more tags like the ones on /programmes, and indeed, how about /programmes/:pipkey:/apml? Or maybe we could provide a service that compares an individual's APML profile to the APML profiles of our radio shows and then generates recommendations appropriately. It seems to me that there could be interesting services built around aggregating or combining APML profiles and lots of opportunities for sites to import APML to bootstrap their recommendations.
* Acknowledgements to .
