Abstract:K-Hub is an efficient high-dimensional data clustering algorithm, but it is sensitive to the choice of initial clustering centers and the instances which besides the class border may not be correctly clustered. In order to solve these problems, an improved method which incorporates active learning and semi-supervised clustering into K-Hub clustering algorithm is proposed. It uses active learning strategy to study pairwise constraints, and then, it uses these pairwise constraints to guide the clustering process of K-Hub. The experiment results demonstrate that the improved method can enhance the performance of K-Hub clustering algorithm.