Music clustering with constraints

This paper studies the problem of building clusters of music tracks in a collection of popular music in the presence of constraints. The constraints come naturally in the context of music applications. For example, constraints can be generated from the background knowledge (e.g., two artists share similar styles) and the user access patterns (e.g., two pieces of music share similar access patterns across multiple users). We present an approach based on the generalized constraint clustering algorithm by incorporating the constraints for grouping music by "similar" artists. The approach is evaluated on a data set consisting of 53 albums covering 41 popular artists. The "correctness" of the clusters generated is tested using artist similarity provided by All Music Guide.