Active Learning Method for Constraint-Based Clustering Algorithms

Semi-supervision clustering aims to improve clustering performance with the help of user-provided side information. The pairwise constraints have become one of the most studied types of side information. According to the previous studies, such constraints increase clustering performance, but the choice of constraints is critical. If the constraints are selected improperly, they may even degrade the clustering performance. In order to solve this problem, researchers proposed some learning methods to actively select most informative pairwise constraints. In this paper, we presents a new active learning method for selecting informative data set, which significantly improves both the Explore phase and the Consolidate phase of the Min-Max algorithm. Experimental results on the data set of UCI Machine Learning Repository, using MPCK-means as the underlying constraint-based semi-supervised clustering algorithm, show that the proposed algorithm has better performance.