Even though time was very limited and many improvements can be made in almost every step the results we obtained are encouraging. The clustering based on the musical features did make (some) sense and the differences between human categorisation and feature based clustering are interesting to say the least. The most striking example to me is that there always was a cluster with (almost) all classical music and a song by Nirvana, which users put in the grunge genre.

It was fun to work on, and I hope there are possibilities to continue this project with more time to make more proper assessments. And as a personal bonus I finally got around to exploring the beautiful programming language Python. Of course I’ve only scratched the surface, but it was enough to make my programmers heart beat a bit faster.

You can read the full paper and if you want, for some reason, get the raw results and source code.