background

The musicexplorer application framework is part of an ongoing research project of the Distributed Computing Group at ETH Zurich. The core innovation consists in a "map of music" that reflects music similarity. The map is a 10-dimensional Euclidean space in which similar tracks are located close to each other, whereas distinct tracks reside at distant locations. Our complete database contains the coordinates of almost half a million tracks. The required notion of music similarity was derived by analyzing the listening patterns of last.fm users.

The applications presented here are proof-of-concept applications that demonstrate the advantages of such a map. The underlying geometric structure, for example, allows to directly visualize entire music collections, and thus to offer intuitive interfaces for similarity based exploration. Smooth playlists can be created by following some path through the map, and the individual taste of a user can quickly be identified by an appropriate decomposition of the space. Due to the compact representation, all these techniques do not require any large databases. Rather, they are directly applicable even in devices such as mobile phones, which are restricted with respect to memory and computing power.

Publications: