Related link: http://www.audioscrobbler.com/index.php
Everyone thinks they’re weird. Especially in music: how can anyone know what I like to listen to? Switching between Bizet’s Carmen, and Waylon Jennings’ Mammas Don’t Let Your Babies Grow Up To Be Cowboys, is confusing even for my taste.
So when someone comes along with a system that tells you not only what you might listen to, but that there are others with similar tastes, well, then we jump at it.
And so it is with Audioscrobbler.com, with which Richard Jones, otherwise known as RJ, might have the answer. In his final year of a computer science degree at Southampton University, RJ is building a system that monitors what you’re listening to, and makes recommendations of new music.
It works like this: You install some software on your machine - currently there are Audioscrobbler plugins for Winamp (Windows), XMMS (Linux) and iTunes (Mac OS X) - and it monitors what you are listening to. This data is sent up to the Audioscrobbler server, which records it against your user name.
After some time, a pattern emerges. You really like bands A, B, C and D, and another user over there likes bands B, C, D and E. So, says the system, chances are you will like band E, and the other user will like band A. Now scale this up to thousands of users, and you have the potential for really good recommendations.
It’s a little more complicated than that, though, says RJ: “People tend to listen to a few artists frequently, and lots of artists occasionally. Seemingly, people have a handful of favourite artists and a load of other ones they like as well. [The system] allows for calculating which songs are most popular in a more ‘natural’ way. For example, three people listening to the same song five times each makes it more popular than one person listening to it 100 times.”
This technique, known as collaborative filtering, is not new. The most well-known project to use this idea was called FireFly. The brainchild of Professor Pattie Maes, from the Massachusetts Institute of Technology, FireFly was also designed to suggest new music. But it asked the users to “rate” different songs or bands. This wasn’t really a weakness at the time - FireFly was launched when dial-up access was the norm, and people couldn’t be expected to be online all the time. But it did mean that their data could be skewed by sociological pressure. People might rate the band by what they think is cool, rather than what they actually listen to.
Audioscrobbler, on the other hand, infers people’s tastes by monitoring what they actually listen to, and not what they say. It is, it must be said, much easier to let the system run in the background, connecting to the net when it needs to, than have to go online specifically, and answer questions.
Amazon, the online retailer, also has a similar system, but it also has limitations. “It is based on what you buy or view online, but I sometimes buy CDs from Amazon as gifts for people, and sometimes buy CDs that I end up not listening to much,” says RJ.
“The Amazon data is a good guideline but I wouldn’t feel comfortable choosing a new CD to buy based on that alone. I hope I’ll be able to trust Audioscrobbler enough to help in my CD-buying decisions in the near future.”
With his final report due in to his supervisor in May, RJ is excited about the potential of the project after graduation. “I’m looking into ways of ‘acoustically fingerprinting’ songs, so that songs can still be identified, even if they are not named correctly_ hopefully Audioscrobbler will become the way to discover and promote new music.”
Meanwhile, however, the popularity of the service - almost 2,000 users - means RJ is looking for help with the hosting.