A Robust k-Means Type Algorithm for Soft Subspace Clustering and Its Application to Text Clustering

Soft subspace clustering are effective clustering techniques for high dimensional datasets. Although several soft subspace clustering algorithms have been developed in recently years, its robustness should be further improved. In this work, a novel soft subspace clustering algorithm RSSKM are proposed. It is based on the incorporation of the alternative distance metric into the framework of kmeans type algorithm for soft subspace clustering and can automatically calculates the feature weights of each cluster in the clustering process. The properties of RSSKM are also investigated. Experiments on real world text datasets are conducted and the results show that RSSKM outperformed some popular clustering algorithms for text mining, while still maintaining efficiency of the k-means clustering process.