Multi-View Clustering via Concept Factorization with
Local Manifold Regularization
Hao Wang, Yan Yang
∗
and Tianrui Li
School of Information Science and Technology
Southwest Jiaotong University, Chengdu 611756, China
∗
Corresponding Author, email: yyang@swjtu.edu.cn
Abstract—Real-world datasets often have representations in
multiple views or come from multiple sources. Exploiting
consistent or complementary information from multi-view data,
multi-view clustering aims to get better clustering quality
rather than relying on the individual view. In this paper, we
propose a novel multi-view clustering method called multi-view
concept clustering based on concept factorization with local
manifold regularization, which drives a common consensus
representation for multiple views. The local manifold regular-
ization is incorporated into concept factorization to preserve
the locally geometrical structure of the data space. Moreover,
the weight of each view is learnt automatically and a co-
normalized approach is designed to make fusion meaningful
in terms of driving the common consensus representation. An
iterative optimization algorithm based on the multiplicative
rules is developed to minimize the objective function. Experi-
mental results on nine reality datasets involving different fields
demonstrate that the proposed method performs better than
several state-of-the-art multi-view clustering methods.
I. INTRODUCTION
Many datasets in real life naturally comprise of different
representations in multiple views or come from multiple
sources, which are called multi-view data. Some examples
are illustrated in Fig. 1. Among these examples, each in-
dividual view suffices for mining knowledge on its own,
but multiple views provide more information which may
improve the performance and quality (e.g., clustering). How-
ever, the main challenge is how to integrate these informa-
tion and give a compatible solution across all views.
Multi-view clustering provides a hopeful way to cluster
data with multiple views, which has attracted more and more
attentions in recent years [1], [2], [3], [4]. Among these
methods, one of the most widely used technique is nonneg-
ative matrix factorization (NMF) [5]. A joint nonnegative
matrix factorization process with the consistency constraint
was formulated in [3], which pushed each view’s solution
towards a common consensus. Besides, one method based
on weighted NMF with L
2,1
regularization was proposed
to learn a latent representation for all views and generate
a consensus matrix in [6]. A feature extraction method via
NMF with local graph regularization for multi-view data
was presented in [7]. Then, the extracted features were used
to cluster the data. However, NMF is not appropriate to
Figure 1. Multi-view data
handle the negative data. As an extension of NMF, concept
factorization (CF) was proposed in [8], which is adapted
to deal with the data containing negative and also is easily
performed in the kernel space.
Besides, both NMF and CF only consider the global
structure of the data space and fail to preserve the locally
geometrical structure. To handle this issue, the graph regu-
larized nonnegative matrix factorization [9] and the locally
consistent concept factorization (LCCF) [10] were proposed
by imposing the manifold regularization on NMF and CF
formulation respectively for single-view data. While, LCCF
does not give a consensus solution for multi-view data.
Additionally, all views were treated equally and the differ-
ence of each view was not considered in such methods [1],
[2], [3], which may degrade the clustering quality. In this
paper, a novel multi-view clustering method called multi-
view concept clustering (MVCC) is proposed. It incorporates
CF, the local manifold regularization and the consistency
constraint into a unified framework, which learns a solution
for each view as well as drives a consensus solution across
all the views. The co-normalized approach is designed
and the weight of each view is respected, which make
the solution of every view compatible and the consensus
meaningful during the factorization process. Besides, an
alternating iterative optimization algorithm is developed to
solve the proposed objective function.
The rest of this paper is organized as follows. In Section
II, a presentation of CF and the local manifold regularization
2016 IEEE 16th International Conference on Data Mining
2374-8486/16 $31.00 © 2016 IEEE
DOI 10.1109/ICDM.2016.34
1245