基于稀疏约束的非负矩阵分解算法研究  

A Research of Non-negative Matrix Factorization Based on Sparse Constraint

作　　者：刘站琳 林小竹[2] 

机构地区：[1]北京化工大学信息科学与技术学院,北京100029 [2]北京石油化工学院信息工程学院,北京102617

出　　处：《北京印刷学院学报》2015年第4期53-58,共6页Journal of Beijing Institute of Graphic Communication

摘　　要：非负矩阵分解是处理高维数据的一种常用方法,对带有稀疏约束的非负矩阵分解算法进行了研究,提出了一种在曼哈顿距离最接近的向量稀疏化算法,并与欧几里得距离最接近的向量稀疏化进行对比,提出的稀疏化算法具有较快的稀疏速度和较好的稀疏效果。实验结果表明,只在非负基矩阵W上加稀疏约束时,得到的非负基矩阵W和非负系数矩阵H的乘积和原非负矩阵V最接近;基于稀疏的非负矩阵分解过程中,选取的迭代步长和迭代次数,会对实验分解结果产生较大影响。Non-negative matrix factorization （ NMF ） is commonly used method for processing high-dimensional data. In this paper,we have a research on NMF with sparse constraints,and propose a sparse vector algorithm have the closest Manhattan distance,compared with the sparse vector algorithm have the closest Euclidean distance,the algorithm has faster speed and better sparse results. Experimental results show that only the non-negative basis matrix W with sparse constraints resulting non-negative basis matrix W multiplied by the non-negative coefficient matrix H is closest to the original non-negative matrix V. In the process of NMF with sparse, iterative step length and the number of iterations selected,will have a greater impact experiments decomposition results.

关 键 词：非负矩阵分解 稀疏约束 曼哈顿距离 欧几里得距离 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]