Semi-supervised clustering via constrained symmetric non-negative matrix factorization

Liping Jing*, Jian Yu, Tieyong ZENG, Yan Zhu

*Corresponding author for this work

Research output: Chapter in book/report/conference proceedingConference contributionpeer-review

4 Citations (Scopus)

Abstract

Semi-supervised clustering based on pairwise constraints has been very active in recent years. The pairwise constraints consist of must-link and cannot-link. Since different types of constraints provide different information, they should be utilized with different strategies in the learning process. In this paper, we investigate the effect of must-link and cannot-link constraints on non-negative matrix factorization (NMF) and show that they play different roles when guiding the factorization procedure. A new semi-supervised NMF model is then proposed with pairwise constraints penalties. Among them, must-link constraints are used to control the distance of the data in the compressed form, and cannot-link constraints are used to control the encoding factor. Meanwhile, the same penalty strategies are applied on symmetric NMF model to handle the similarity matrix. The proposed two models are implemented by an alternating nonnegative least squares algorithm. We examine the performance of our models on series of real similarity data, and compare them with state-of-the-art, illustrating that the new models provide superior clustering performance.

Original languageEnglish
Title of host publicationBrain Informatics - International Conference, BI 2012, Proceedings
Pages309-319
Number of pages11
DOIs
Publication statusPublished - 2012
Event2012 International Conference on Brain Informatics, BI 2012 - Macau, China
Duration: 4 Dec 20127 Dec 2012

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume7670 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference2012 International Conference on Brain Informatics, BI 2012
Country/TerritoryChina
CityMacau
Period4/12/127/12/12

Scopus Subject Areas

  • Theoretical Computer Science
  • Computer Science(all)

User-Defined Keywords

  • NMF
  • Pairwise Constraints
  • Semi-supervised Clustering
  • Symmetric NMF

Fingerprint

Dive into the research topics of 'Semi-supervised clustering via constrained symmetric non-negative matrix factorization'. Together they form a unique fingerprint.

Cite this