Research And Application Of Multi-view Based Unsupervised Subspace Clustering Algorithm

With the advancement of information acquisition technology,multi-view data becomes ubiquitous.We had easier access to the data with multi-view representations from heterogeneous feature space.Multi-view clustering aims to discover the inherent structure by using complementary and consensus information across multiple views.It has broad application scenario,such as computer vision,natural language processing,social multimedia,and so on.Thus,multi-view clustering has become more and more popular in machine learning and data mining fields.Although many multi-view clustering algorithms have been proposed recently and have achieved good results in specific area,existing algorithms still have some shortcomings.Multi-view data is often unlabeled,partial or full of noise.This unique challenges and properties motivate us to propose a joint feature selection and self-representation algorithm(JFSSR),and extend it to a robust multi-view subspace clustering algorithm(RMSC),which learns a consensus affinity matrix with the ideal subspace structure.Concretely,RMSC learns the consensus graph across diverse views with exactly k connected components(k is the number of clusters),which is encoded by a block diagonal self-representation matrix.Besides,we emphasize L_2,1-norm minimization on the loss function to reduce redundant and irrelevant features,and implicitly assign an adaptive weight to each view without introducing additional parameters.Lastly,an alternating optimization algorithm is derived to solve the nonconvex formulated objective.Extensive empirical results on both synthetic data and real-world benchmark data sets show that RMSC consistently outperforms several representative multi-view clustering approaches.Multi-view data does not necessarily conform to the linear subspace models,and most existing multi-view clustering methods only consider the consistency or the diversity of different views.This unique challenges and properties motivate us to propose a deep multi-view subspace clustering algorithm(DMVSC).This architecture is built upon deep auto-encoders,which non-linearly maps the multi-view data into a set of latent spaces.More importantly,we introduce a self-expressive layer to learn a shared consistency representation of all views and a set of specific representations for each view.Specifically,consistency representation models the common properties among all views,while specificity captures the inherent difference in each view.Empirical results on four benchmark datasets demonstrate that the proposed approach achieves better performance over several state-of-the-arts.By combining algorithm research with practical applications,we design and implement an analysis system based on clustering methods.The system mainly includes three parts,which are the main interface module,digital visualization module and clustering algorithms module.With this cluster analysis system,you can understand the experimental data set intuitively,and perform related clustering algorithms and visualize the results.