Clustering with Constraints

The area of clustering with constraints makes use of hints or advice in the form of constraints to aid or bias the clustering process. The most prevalent form of advice are conjunctions of pair-wise instance level constraints of the form must-link (ML) and cannot-link (CL) which state that pairs of instances should be in the same or different clusters respectively. Given a set of points P to cluster and a set of constraints C, the aim of clustering with constraints is to use the constraints to improve the clustering results. Constraints have so far being used in two main ways: a) Writing algorithms that use a standard distance metric but attempt to satisfy all or as many constraints as possible and b) Using the constraints to learn a distance function that is then used in the clustering algorithm.