Semi-supervised Parameter-Free Divisive Hierarchical Clustering of Categorical Data

Semi-supervised clustering can yield considerable improvement over unsupervised clustering. Most existing semi-supervised clustering algorithms are non-hierarchical, derived from the k-means algorithm and designed for analyzing numeric data. Clustering categorical data is a challenging issue due to the lack of inherently meaningful similarity measure, and semi-supervised clustering in the categorical domain remains untouched. In this paper, we propose a novel semi-supervised divisive hierarchical algorithm for categorical data. Our algorithm is parameter-free, fully automatic and effective in taking advantage of instance-level constraint background knowledge to improve the quality of the resultant dendrogram. Experiments on real-life data demonstrate the promising performance of our algorithm.