Applying Constrained Clustering for Active Exploration of Music Collections

In this paper we investigate the capabilities of constrained clustering in application to active exploration of music collections. Constrained clustering has been developed to improve clustering methods through pairwise constraints. Although these constraints are received as queries from a noiseless oracle, most of the methods involve a random procedure stage to decide which elements are presented to the oracle. In this work we apply spectral clustering with constraints to a music dataset, where the queries for constraints are selected in a deterministic way through outlier identification perspective. We simulate the constraints through the ground-truth music genre labels. The results show that constrained clustering with deterministic outlier identification method achieves reasonable and stable results through the increment of the number of constraint queries.