基于正交非负矩阵分解的K-means聚类算法研究

Orthogonal Non-negative Matrix Factorization for K-means Clustering

为提高K-means聚类算法在高维数据下的聚类效果,提出了一种基于正交非负矩阵分解的K-means聚类算法。该算法对原始数据进行非负矩阵分解,并分别通过改进的Gram-Schmidt正交化和Householder正交化加入了正交约束,以保证低维特征的非负性,增加数据原型矩阵的正交性,然后进行K-means聚类。实验结果表明,基于IGSONMF和H-ONMF的K-means聚类算法在处理高维数据上具有更好的聚类效果。

The orthogonal NMF K-means clustering algorithm based on basic theory of NMF was proposed to improve the quality of K-means clustering in high-dimensional data.We presented orthogonal NMF algorithm,added orthogonal restraint to data prototype matrix from factorization with improved Gram-Schmidt and Householder orthogonalization separately,which both ensure non-negative of low-dimensional feature and enhance the orthogonality of matrix,and then made K-means clustering.Experimental results show that K-means clustering based on H-ONMF has better clustering results on high-dimensional data.