Kappa Based Weighted Multi-View Clustering with Feature Selection

Abstract

In recent years, multi-view clustering has been developed to a high level and widely used in many real-world applications. Since different views are variable representations of the same instance set, thus weighted multi-view clustering with feature selection (WMCFS) has been proposed to use information from multiple views simultaneously to boost the clustering results. WMCFS not only combines information from multiple views but also performs feature selection so as to solve high-dimensional data sets. Although related experiments validate the effectiveness of WMCFS, due to kappa is an index to measure the inter-rater agreement for qualitative (categorical) items, thus we introduce kappa to WMCFS and propose a kappa based WMCFS (KWMCFS) to boost the clustering performance further. Experiments on multi-view data sets Mfeat, Reuters, and Corel validate that compared with WMCFS, introducing kappa boosts the clustering and classification performances.