Multi-view Clustering Algorithm Based on Variable Weight and MKL

摘要

Compared with Single-view clustering, Multi-view clustering analysis exploits more hidden information. Multiple kernel learning (MKL) performs its superiority in heterogeneous sources and solves the problem of selection of kernel functions. Many existing multi-view literatures based on MKL consider instances in each view equally and overlook the difference among them. In this paper, a multi-view clustering algorithm based on variable weight and MKL (called MVMKC) is proposed. MVMKC improves clustering quality with more-refined analyses on data. To be specific, it uses an improved weighted Gaussian kernel rather than the traditional combined kernel function. Meanwhile, variable weights are introduced to measure the contribution of instance in different views. Experimental results on real-world datasets demonstrate the effectiveness of the proposed approach.