View site View screenshots
Pic

Music Recommendation Software, Started Winter 2006, Ongoing

Components :

  • Java
  • Server API (PHP)
  • MySQL

TrackTaste

This is an extension to my masters project on music recommendation. You can read the project report here.

I was never really happy with the way the project turned out and know I can make the algorithm much faster and more efficient. Whenever I have free time I like to go back to this project to try and tweak it to be better, with the hope being that I will eventually release it as a open source project.

The idea was that with current music recommendation it's difficult to see where the recommendation came from. Take iTunes for example, it will recommend songs to you, but you don't know whether they're just forcing the popular songs on your, or whether they're based on the things you've bought/listened to.

So I created the "Visualizer". It shows you as a graph how the recommendation was made, using the JUNG graphing framework for Java. But more than that, it allows you to manipulate the graph in real-time in order to adjust the recommendations. So you can adjust the similarity ratings between yourself and other users and watch the graph change to reveal new recommendations.

But before I could create the visualizer I needed to create a recommendation system. I created a collaborative filtering based music recommendation engine, and over several months asked users to listen to music and the system would keep track of what they listened to, and send this information to me, and evetually it would recommend music so that they users could use the visualizer.

This is an ongoing project.