Two algorithms for orthogonal nonnegative matrix factorization with application to clustering

被引:125
|
作者
Pompili, Filippo [1 ]
Gillis, Nicolas [2 ]
Absil, P. -A. [3 ]
Glineur, Francois [3 ,4 ]
机构
[1] Univ Perugia, Dept Elect & Informat Engn, I-06100 Perugia, Italy
[2] Univ Mons, Dept Math & Operat Res, B-7000 Mons, Belgium
[3] Catholic Univ Louvain, ICTEAM Inst, B-1348 Louvain, Belgium
[4] Catholic Univ Louvain, CORE, B-1348 Louvain, Belgium
关键词
Nonnegative matrix factorization; Orthogonality; Clustering; Document classification; Hyperspectral images;
D O I
10.1016/j.neucom.2014.02.018
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Approximate matrix factorization techniques with both nonnegativity and orthogonality constraints, referred to as orthogonal nonnegative matrix factorization (ONMF), have been recently introduced and shown to work remarkably well for clustering tasks such as document classification. In this paper, we introduce two new methods to solve ONMF. First, we show mathematical equivalence between ONMF and a weighted variant of spherical k-means, from which we derive our first method, a simple EM-like algorithm. This also allows us to determine when ONMF should be preferred to k-means and spherical k-means. Our second method is based on an augmented Lagrangian approach. Standard ONMF algorithms typically enforce nonnegativity for their iterates while trying to achieve orthogonality at the limit (e.g., using a proper penalization term or a suitably chosen search direction). Our method works the opposite way: orthogonality is strictly imposed at each step while nonnegativity is asymptotically obtained, using a quadratic penalty. Finally, we show that the two proposed approaches compare favorably with standard ONMF algorithms on synthetic, text and image data sets. (C) 2014 Elsevier B.V. All rights reserved.
引用
收藏
页码:15 / 25
页数:11
相关论文
共 50 条
  • [21] Document clustering using nonnegative matrix factorization/
    Shahnaz, F
    Berry, MW
    Pauca, VP
    Plemmons, RJ
    INFORMATION PROCESSING & MANAGEMENT, 2006, 42 (02) : 373 - 386
  • [22] Nonnegative Matrix Factorization for Document Clustering: A Survey
    Hosseini-Asl, Ehsan
    Zurada, Jacek M.
    ARTIFICIAL INTELLIGENCE AND SOFT COMPUTING, ICAISC 2014, PT II, 2014, 8468 : 726 - 737
  • [23] Structure constrained nonnegative matrix factorization for pattern clustering and classification
    Lu, Na
    Miao, Hongyu
    NEUROCOMPUTING, 2016, 171 : 400 - 411
  • [24] Regularized asymmetric nonnegative matrix factorization for clustering in directed networks
    Tosyali, Ali
    Kim, Jinho
    Choi, Jeongsub
    Jeong, Myong K.
    PATTERN RECOGNITION LETTERS, 2019, 125 : 750 - 757
  • [25] Nonnegative Matrix Factorization: Algorithms, Complexity and Applications
    Moitra, Ankur
    PROCEEDINGS OF THE 2015 ACM ON INTERNATIONAL SYMPOSIUM ON SYMBOLIC AND ALGEBRAIC COMPUTATION (ISSAC'15), 2015, : 15 - 16
  • [26] Algorithms and applications for approximate nonnegative matrix factorization
    Berry, Michael W.
    Browne, Murray
    Langville, Amy N.
    Pauca, V. Paul
    Plemmons, Robert J.
    COMPUTATIONAL STATISTICS & DATA ANALYSIS, 2007, 52 (01) : 155 - 173
  • [27] The rise of nonnegative matrix factorization: Algorithms and applications
    Guo, Yi-Ting
    Li, Qin-Qin
    Liang, Chun-Sheng
    INFORMATION SYSTEMS, 2024, 123
  • [28] Assessment of nonnegative matrix factorization algorithms for electroencephalography spectral analysis
    Hu, Guoqiang
    Zhou, Tianyi
    Luo, Siwen
    Mahini, Reza
    Xu, Jing
    Chang, Yi
    Cong, Fengyu
    BIOMEDICAL ENGINEERING ONLINE, 2020, 19 (01)
  • [29] Assessment of nonnegative matrix factorization algorithms for electroencephalography spectral analysis
    Guoqiang Hu
    Tianyi Zhou
    Siwen Luo
    Reza Mahini
    Jing Xu
    Yi Chang
    Fengyu Cong
    BioMedical Engineering OnLine, 19
  • [30] RANDOMIZED ALGORITHMS FOR SYMMETRIC NONNEGATIVE MATRIX FACTORIZATION
    Hayashi, Koby
    Aksoy, Sinan g.
    Ballard, Grey
    Park, Haesun
    SIAM JOURNAL ON MATRIX ANALYSIS AND APPLICATIONS, 2025, 46 (01) : 584 - 625