Articles review on the net, revue d'articles sur la toile


Inscription : feeds, flux :
(Atom) Gabriel Real World News

20 août 2007

Filtering the music in your life

A problem with the rapid increase in our ability to store data, is not only that we might have trouble finding things but that it could be difficult just deciding what to look for.

As we collect thousands of digital music tracks or videos, how will we ever be able to decide what to watch or listen to? The choice becomes overwhelming. Or at least that's the thinking behind a recommendation engine developed by Bath-based physicist Martin Hopkins.

His idea came from trying to organise his own digital music collection and compile playlists. He says: "We find everyone has the Beatles in their collection so that doesn't really help us much , but if you have Frank Zappa or Captain Beefheart we'll be able to tell exactly what mood you're in." The search software spawned from the idea, and called The Filter, draws on comparatively ancient mathematics ; the statistical probability theorem invented by 18th Century churchman Reverend Thomas Bayes.

Mr Hopkins says Bayesian probability is the ideal mathematical model for this kind of search "It mimics how we come to a conclusion in life. It gathers evidence from a variety of sources and gives them weighting. From this we can work out the probability of an outcome." The software looks at the music you have and compiles playlists based on the relationships between tracks. "It's a software download that sits of your computer - it goes through every music track on your computer, and that information is sent anonymously back to our sever. From that we can create a whole series of cross references." The software is not foolproof, however.

On The Filter's website there is a facility to test it out and Digital Planet presenter Gareth Mitchell entered Stone Roses, Razorlight and Jethro Tull, as his favourite bands - but was startled to find that one of the recommendations was Busted. Mr Hopkins says the algorithm only has the data in the evidence base to work from, but that half the fun is trying to break the system.

"You can put in Nancy Sinatra and the Sex Pistols and see what you get. There may not be a relationship, in which case it will confuse the software." Despite the fun or annoyance of using this system to lazily suggest music, the software behind it may have a much wider use. Mr Hopkins says it could be applied to any type of data "The whole approach we have is content agnostic, we don't analyse the attributes of the content. It's really the relationships between pieces of content, so it works for movies, it works for books."

The software is being backed by musician Peter Gabriel, who describes it as being part of the second wave of the digital revolution; not offering freedom of choice, but freedom from choice.

Aucun commentaire: