A Kernel-Learning Approach to Semi-supervised Clustering with Relative Distance Comparisons