Research On New Multi-view Clustering Algorithm

At present,multi-view clustering has made important progress,but it still faces many challenges.One of the key challenges is: Although most multi-view clustering methods propose different perspective coordination mechanisms,most of them are performing multi-view collaboration.Only using visual original spatial information or only some hidden spatial information,thus failing to fully realize the synergy between visual and hidden information.In addition,the topological structure is important information of the data set,but most multi-view clustering algorithms fail to make full use of such information for collaboration when they perform collaborative learning.In response to the above challenges,this paper proposes two multi-view clustering algorithms.Compared with the existing multi-view learning algorithms,the proposed algorithm has a significant improvement in theory and performance.The following are the two main tasks of this article for multi-view clustering:1)The first work is to propose a multi-view fuzzy clustering algorithm that takes into account both visual and hidden information and feature weighting.This algorithm realizes collaborative learning from each view under the framework of fuzzy clustering.On the one hand,the personalized information is obtained by clustering the weights of the features in each view.On the other hand,the algorithm extracts the coefficient matrix shared by the multi-view data set in the way of feature learning,obtains common information,and realizes a kind of multi-view learning with the collaboration of visual and hidden views.2)The second work is to propose a collaboratively enhanced multi-view clustering that takes into account double hidden views and network lasso constraints.The algorithm first introduces two different hidden views to characterize data.One is to use adaptive non-negative matrix factorization technology to construct a common hidden feature view of multiple display views,and the other is to use the sample's membership of different clusters.The degree information is used to construct another common hidden view of each visual view,and then a multi-view fuzzy clustering algorithm with the collaboration of visual and hidden views and the collaborative enhancement of spatial topological information is proposed.