Learning a Metric during Hierarchical Clustering based on Constraints.
Korinna Bade, Andreas Nürnberger · 2009
Constrained clustering has many useful applica-tions. In this paper, we consider applications, in which a hierarchical target structure is prefer-able. Therefore, we constrain a hierarchical ag-glomerative clustering through the use of MLB constraints, which provide information about hi-erarchical relations between objects. We pro-pose an algorithm that learns a suitable metric according to the constraint set by modifying the metric whenever the clustering process violates a constraint. We furthermore combine this ap-proach with an instance-based constrained clus-tering to further improve the cluster quality. Both approaches have proven to be very successful in a semi-supervised setting, in which constraints do not cover all existing clusters.