| In recent years,with the development and progress of mobile Internet technology,people have increasing ways to get information,so the data structure obtained is more and more complex.Facing high-dimensional data,the traditional clustering algorithm cannot achieve the ideal clustering effect.Subspace clustering is widely studied and applied as an effective algorithm to deal with high-dimensional data clustering.Low rank representation and sparse subspace clustering are two typical subspace clustering algorithms.However,the above clustering algorithm is only for the information of a single view,and cannot play an accurate clustering effect on the complex and diverse data in real life.Therefore,multi-view subspace clustering algorithm has attracted wide attention.However,the multi-view subspace clustering algorithm directly reconstructs data points on the original view,and generates the individual subspace for each view.However,each single view alone is usually not sufficient to describe data points,which makes the data reconstruction by using only one view ineffective.Therefore,latent multi-view subspace clustering algorithm is proposed,the algorithm clusters data points with latent representation and simultaneously explores underlying complementary information from multiple views.The algorithm learns the underlying latent representation,it reconstructs the data based on the learned latent representation,and the obtained latent representation can describe the data points more comprehensively.However,the latent representation learned in the latent multi-view subspace clustering algorithm is obtained through the data projection on each view,which is lack of guidance.Therefore,to overcome this limitation,this thesis proposes latent representation guided multi-view subspace clustering algorithm.Through the Laplace matrix of the Laplace term introduced,the algorithm explores the relationship between the various view data introduced in advance to construct the guide information.Based on this guidance information,a better potential representation matrix is built to make the clustering effect more accurate.In addition,in this thesis,the sensitivity of parameters in the proposed algorithm was analyzed,and the influence of parameter variation on clustering performance is studied.Finally,numerical experiments are carried out on three multi-view data sets,and the experimental results show that the latent representation guided multi-view subspace clustering algorithm proposed in this thesis has superior clustering performance. |
