不完备多视图的在线反向图正则化聚类

Online Reverse Graph Regularized Clustering for Incomplete Multi-view

在如今的大数据时代,多视图数据引起了越来越多的关注,对多视图聚类的假设是所有视图都是完整的,然而,这种假设在实际应用中很难得到满足。因此不完备多视图聚类是一个重要挑战。针对大规模的不完备多视图数据,考虑到其数据的特征,利用互补性和一致性,论文提出了一种基于非负矩阵分解的在线反向图正则化聚类方法,首先利用加权非负矩阵分解作为基础模型,考虑到缺失实例的影响,引入一个动态权重矩阵;其次,学习所有视图的潜在特征矩阵并得到一个共识矩阵;同时,考虑到挖掘数据的局部结构,在基础模型上增加反向图正则化项;最后,对于大规模的数据,分块处理多视图数据以减少内存需求。在四个真实的数据集上进行了大量实验证明了所提出的方法的有效性。

In today's era of big data,multi-view data has attracted more and more attention.The assumption of multi-view clustering is that all views are complete.However,this assumption is difficult to meet in practical application.Therefore,incomplete multi-view clustering is an important challenge.For large-scale incomplete multi-view data,considering the characteristics of the data,using complementarity and consistency,this paper proposes an online reverse graph regularization clustering method based on non-negative matrix factorization.Firstly,weighted non negative matrix factorization is used as the basic model,and a dynamic weight matrix is introduced considering the influence of missing examples.Secondly,the potential feature moments of all views are learned.At the same time,considering the local structure of the mining data,the regularization term of reverse graph is added to the basic model.Finally,for large-scale data,multi view data is processed in blocks to reduce the memory requirement.Experiments on four real datasets show the effectiveness of the proposed method.