【摘要】 针对高维海量数据，现有的多视图聚类方法存在无法发现高维视图隐藏信息、聚类效果差等问题。结合均方残差(Mean Squared Residue, MSR)思想，提出了一种基于非负矩阵分解的均方残差多视图聚类方法(Mean Squared Residue Non-negative Matrix Factorization, MSRNMF)。首先，采用改进的非负矩阵分解方法结合流形学习、希尔伯特-施密特独立性准则计算各单视图的系数矩阵，不仅降低了多视图中各个视图的维度，而且有效地提取了高维数据中的隐藏信息；其次，采用谱聚类算法对各单视图的系数矩阵进行聚类，获得单视图聚类簇；接着，利用均方残差思想，针对各单视图聚类结果进行融合，得到最终多视图聚类结果；最后，以标准数据集和古建数据集为对象进行验证，实验结果表明该算法在精度上优于MVCF,GPSNMF,GPMVNMF,DMF和MCLES,在古建筑集上效果明显，进而验证了算法的有效性。更多还原

【Abstract】 For high-dimensional massive data, the existing multi-view clustering methods have some problems, such as failing to discover the hidden information of high-dimensional view and poor clustering effect. With the idea of mean square residuals(MSR),a method of clustering with mean squared residue based on non-negative matrix factorization(MSRNMF) is proposed. Firstly, the improved non-negative matrix factorization method combined with manifold learning and Hilbert-Schmidt independence criterion is used to calculate the coefficient matrix of each single view, which not only reduces the dimensions of each view in the multi-view, but also effectively extracts the hidden information in the high-dimensional data. Secondly, spectral clustering algorithm is used to cluster the coefficient matrix of each single view, and the single view cluster is obtained. Then using the idea of mean square residual, the clustering results of each single view are fused to obtain the final multi-view clustering results. Finally, standard data sets and ancient construction data sets are used for verification. The experimental results show that the accuracy of the proposed algorithm is better than that of MVCF,GPSNMF,GPMVNMF,DMF and MCLES,and the effectiveness of it is verified.更多还原