Advances in constrained clustering

Abstract

Constrained clustering (semi-supervised learning) techniques have attracted more attention in recent years. However, the commonly used constraints are restricted to the instance level, thus we introduced two new classifications for the type of constraints: decision constraints and non-decision constraints. We implemented applications involving non-decision constraints to find alternative clusterings. Due to the fact that randomly generated constraints might adversely impact the performance, we discussed the main reasons for carefully generating a subset of useful constraints, and defined two basic questions on how to generate useful constraints.