基于超图正则NMF的自适应半监督多视图聚类

Adaptive Semi-supervised Multi-view Clustering Based on Hypergraph Regular NMF

图正则非负矩阵分解(GNMF)虽然已成为大量多视图聚类方法的基本框架,但其尝试用简单图融合来自不同视图的复杂数据关系,同时获得一致性判别表示,这无疑有很大挑战性。为了更好地应对多视图数据的聚类任务,本文提出一种基于超图正则非负矩阵分解的半监督多视图聚类方法ASMCHNMF。该方法通过构造超图,学习来自多个视图的数据高阶关系,为合理利用现实世界中可获取的标签信息,引入标签约束项进行半监督学习。此外,该方法同时考虑一致性信息和互补性信息的学习,采用自适应措施区分不同视图的贡献,并使用交替迭代算法来对主函数进行优化。在7个真实数据集上的对比实验表明,在其中6个数据集上,ASMCHNMF算法的ACC和NMI指标均优于经典算法和当前先进算法。

Although graph regularized non-negative matrix factorization(GNMF)has become the basic framework for a large number of multi-view clustering methods,it is undoubtedly a great challenge to fuse complex data relationships from different views with a simple graph and obtain a consistent discriminative representation at the same time.In order to better deal with the clustering task of multi-view data,a semi-supervised multi-view clustering method based on hypergraph regularized non-negative matrix factorization is proposed.Specifically,by constructing a hypergraph,this method learns the high-order relationships of data from multiple views.In order to make reasonable use of the label information available in the real world,the label constraint is introduced for semi-supervised learning.In addition,this method considers the learning of consistency information and complementarity information at the same time,adopts adaptive measures to distinguish the contributions of different views,and uses an alternating iterative algorithm to optimize the objective function.The comparative experimental results on 7 real datasets show that the proposed algorithm is superior to other classical algorithms and current advanced algorithms in ACC and NMI indicators on 6 datasets.