Multi-View Clustering Based On Integrated Weight Learning And Non-Negative Matrix Factorization

Recently,with the rapid development of the Internet and the gradual explosion of information,technology based on the analysis of massive internet data has been increasingly valued by industry and academia.Tasks based on the identification,analysis,classification,clustering,and association mining of Internet data are increasingly becoming an important part of Internet information activities.At the same time,due to its sparsity,complexity,and weak correlation,Internet data poses severe requirements for related Internet information mining technologies and underlying algorithms.Many traditional data analysis methods in the past cannot meet the requirements;especially for some basic data analysis methods(classification,clustering,correlation analysis,etc.).At the same time,Internet data has a specific feature—multi-view,that is,a single data has multiple interpretations.The processing and model of this type of data came into being.In this paper,a new clustering algorithm based on multi-view clustering is proposed,which combines non-negative matrix decomposition method,genetic algorithm competition model and neural network method,and applies this technology to conventional multi-view data clustering.Firstly,this article will introduce the basic knowledge and prior knowledge in one or two chapters.Includes,but is not limited to,existing research in multi-view clustering methods in current academia and industry,the basic concepts of clustering algorithms and the introduction of some traditional clustering algorithms.This section clarifies the writing intention and the article planning by introducing the basic and preliminary work of this article in two chapters.Secondly,this paper introduces the knowledge of multi-view clustering and non-negative matrix factorization.Multi-view clustering is a clustering task model for multi-view data,which can show high performance when it is oriented to multi-view data processing.Non-negative matrix factorization is a matrix decomposition technique.Through non-negative matrix factorization,feature reduction,feature extraction,and data compression can be performed on the correlation matrix.Finally,it is the focus of this paper.Combining multi-view clustering decomposition method,non-negative matrix decomposition method and view synthesis method,a new clustering algorithm based on multi-view clustering is proposed.In this part,this paper used non-negative matrix to extract the image features of the subview,and used the matrix method,the competition algorithm and the neural network method to synthesize and normalize the target multi-view features.Through the above systematic discussion and research,and sufficient experiments at the same time,the three models are compared horizontally and vertically with two different types of datasets.It shows that the integrated clustering method based on non-negative matrix factorization(MNMF),multi-view integrated clustering method based on competitive genetic algorithm(CMC)and multi-view clustering method based on neural network integration(NMC)are superior to traditional clustering algorithm in many performance indicators.