结构正则化多视图非负矩阵分解

现存非负矩阵分解（non-negative matrix factorization，NMF）研究多考虑单一视图分解数据，忽略了数据信息的全面性。此外，NMF限制其获取数据的内在几何结构。针对以上问题，提出一个结构正则化多视图非负矩阵分解算法（structure regularized multi-view nonnegative matrix factorization，SRMNMF）。首先，通过主成分分析来对数据进行全局结构的判别式学习；其次，利用流形学习来捕获数据的局部结构；然后，通过利用多视图数据的多样性和差异性来学习表征。模型提升了算法聚类的整体性能，更加有效地挖掘数据的结构信息。此外，采用高效的交替迭代算法优化目标函数得到最优的因子矩阵。在六个数据集上与现存的代表性方法比较，所提出的SRMNMF的准确率、NMI和Purity分别最大提高4.4%、6.1%和4.05%。

Structure regularized multi-view nonnegative matrix factorization

Existing Non-negative Matrix Factorization(NMF) studies mostly consider a single view to decompose data, ignoring the comprehensiveness of data information. Additionally, NMF limit their access to the intrinsic geometry of the data. This paper proposed a novel matrix factorization method, called structure regularized multi-view nonnegative matrix factorization(SRMNMF). Specifically, this paper first perform discriminative learning of the global structure of the data through principal component analysis. Then, it used manifold learning to capture the local structure of the data. Finally, it learnt representations by exploiting the diversity and difference of multi-view data. The model improved the overall performance of the algorithm clustering and mined the structural information of the data more effectively. The objective function of SRMNMF could be easily optimized using an efficient alternate iterative algorithm. Comparing with existing representative methods on 6 datasets, the proposed SRMNMF achieves a maximum improvement of 4.4%, 6.1% and 4.05% in accuracy, NMI and Purity, respectively.