Incomplete Multi-view Data Clustering Analysis

With the rapid development of data acquisition technology and data extraction technology,a large amount of multi-view data has been obtained in the real world.In general,different view data can provide complementary information to each other,which is very helpful for practical applications.Therefore,multi-view data analysis is widely used in data mining,information retrieval and multimedia fields.Multi-view clustering is one of the representative algorithms in the field of multi-view learning.In practical applications,due to the failure of the data acquisition device or the random noise during the transmission process,data missing in a certain view part often occurs,which directly leads to the rapid decline of the performance of the existing multi-view data clustering algorithm.For such data,the existing incomplete multi-view clustering methods are proposed.Most of these algorithms complement the subspace or coefficient matrix of missing data by linear mapping,and then divide the multi-view data into different clusters.These methods neglect the spatial structure between views,which leads to the extraction of nonlinear features can not better describe the complementary information of multi-view data;without considering the diversity and authenticity of the complementary vision,the clustering effect is not good;In view of the above problems,this paper analyzes the multi-view clustering algorithm and the lack of completion algorithm to improve the algorithm from different angles.The main contents of this paper are as follows:(1)Aiming at the shortcomings of the existing incomplete multi-view clustering algorithmsuch as waste of information,large amount of calculation,complete diversity of vision and poor authenticity,based on the Generative Adversarial Nets(GAN),this paper makes full usof the complementarity of multi-view data to learn the relationship between different viewsand proposes a GAN-based incomplete multi-view clustering algorithm,namely GP-MVCWhen there is a large proportion of some views data in multi-view data sets,the algorithm can still repair the data and obtain more reliable clustering results.In this paper,the MNISTHW and BDGP three commonly used multi-view data sets have been fully tested with thmissing ratios of 0.1,0.3,0.5,0.7,0.9.The results show that the proposed method improvethe clustering of missing multi-view data performance.(2)GP-MVC only considers the characteristics of difference and complementarity between different view data,and does not consider the consistency information between different view.In order to further improve the performance of the incomplete multi-view data clustering algorithm,this paper proposes an incomplete multi-view clustering algorithm based on consistency Generative Adversarial Nets,namely CGP-MVC.The algorithm can learn a shared latent representation for incomplete multi-view data,not only to capture consistent information for different views of clustering,but also to repair missing data.Experimental results on multiple data sets demonstrate the superiority of the proposed algorithm.