Research On Multi-view Clustering Algorithms

Clustering is a basic technology in data mining,computer vision,pattern recognition,and biomedical information processing,and has been widely used for face recognition and recommendation.In recent years,with the rapid development of the Internet,social networks and big data,data can be collected from different sources or described from different views.For example,a picture can be represented by its color or texture features.Due to the emergence of such multi-view data,multi-view clustering has attracted more and more attention.In recent years,many multi-view clustering methods have been proposed,among which the performance of multi-view subspace clustering is remarkable,especially for highdimensional data.However,the existing multi-view subspace clustering methods are generally divided into two stages,learning subspace representation and clustering,that is,the representation learning and clustering cannot be performed simultaneously.In this paper,a new multi-view subspace clustering method is proposed,which integrates the two stages into a unified framework,learning the representation and clustering of each view simultaneous,meanwhile,a pairwise co-regularized term is proposed to ensure the consistency across views.Another popular multi-view clustering method is multiple kernel learning.The kernels in multiple kernel learning naturally correspond to multiple views,and proper combination of kernels can improve clustering performance.The key to multiple kernel learning is to assign a reasonable weight to each kernel to combine an optimal consensus kernel.But multiple kernel learning inevitably introduces multiple weight parameters,and improper weight assignment may even result in worse clustering performance than single kernel learning.In this paper,a new multiple kernel k-means clustering method with a self-weighted learning approach.Unlike other multiple kernel methods in the literature,this method can learn the consensus kernel and automatically assign the optimal weight to each kernel without introducing other parameters.In this paper,two multi-view clustering algorithms are proposed,co-regularized multiview subspace clustering algorithm and a self-weighted multiple kernel k-means clustering method.The experiments on several real-world multi-view datasets demonstrate the effectiveness and superiority of proposed methods.