Exploration of Different Constraints and Query Methods with Kernel-based Semi-supervised Clustering

Semi-supervised clustering makes use of a small amount of supervised data to aid unsupervised learning. The method used to obtain the supervised information, and the way such information is integrated within the learning algorithm can greatly affect the final result. This paper introduces two different kernel-based semi-supervised clustering algorithms, and investigates the power of kernel methods in principle. Moreover, driven by practice, two methods to obtain supervised data are considered. We compare our kernel-based semi-supervised clustering approaches with semi-supervised K-means and unsupervised kernel K-means. The experimental results show that both our methods can outperform the others, regardless of the technique used to generate the supervised data.