ORIGINAL PAPER
A multi-granular network representation learning method
Jie Chen
1,2
•
Ziwei Du
1,2
•
Xian Sun
1,2
•
Shu Zhao
1,2
•
Yanping Zhang
1,2
Received: 28 May 2019 / Accepted: 16 August 2019
Ó Springer Nature Switzerland AG 2019
Abstract
Granular computing (GrC) as a problem-solving concept and new inf ormation processing paradigm is deeply rooted in
human thinking, which has attracted many researchers to study it theoretically, and has gradually applied to data-driven
problems. Network embedding, as known as network representation learning, aiming to map nodes in network into a low-
dimensional representation, is a data-driven problem. Most existing methods are based on a single granular, which learn
representations from local structure of nodes. But global structure is important information on the network and has been
proven to facil itate several network analysis tasks. Therefore, how to introduce GrC into network embedding to obtain a
multi-granular network representation that preserves the global and local structure of nodes is a meaningful and tough task.
In this paper, we introduce Quotient Space Theory, one of the GrC theories into network embedding and propose a Multi-
Granular Network Representation Learning method based on Quotient Space Theory (MG_NRL, for short), which can
preserve global and local structure at different granularities. Firstly, we granulate the network repeatedly to obtain a multi-
granular network. Secondly, the embedding of the coarsest network is com puted using any existing embedding method.
Finally, the networ k representation of each granular layer is learned by recursively refining method from the coarsest
network to original network. Experimental results on multi-label classification task demonstrate that MG_NRL signifi-
cantly outperforms other state-of-the-art methods.
Keywords Granular computing Quotient space theory Network embedding Multi-granular Multi-label classificatio n
1 Introduction
Granular computing (GrC) is originally called information
granularity or information granulation related to fuzzy sets
research (Zadeh 1997). It is generally accepted that 1997 is
regarded as the birth of GrC. In recent years, the basic
notions and principles of GrC occurred under various forms
in many disciplines and fields (Zadeh 1997; Yao 2001). We
have witnessed a rapid development and fast growing
interest in this topic (Bargiela and Pedrycz 2008; Qian
et al. 2010; Wang 2017; Zhao et al. 2017; Wang et al.
2017a). In the existing Granular Computing theories, it is
well accepted that the theories of fuzzy sets (Zadeh 1965)
and rough sets (Pawlak 1982) are two primary contribu-
tions that have occurred from the emergence of GrC (Lin
1999, 2003). Inspired by Hobbs’ idea that the world can be
dealt with hierarchically (Hobbs 1990), Zhang and Zhang
(1990, 1992, 2007) propose Quotient Space Theory, which
is one of the theoretical research in GrC theories. The core
idea of the quotient space theory is forming a suit-
able granular space to describe and solve the problem, as
well as constructing the relation of different granular
spaces to simpl ify the problem.
Multi-granular as an intrinsic logic requirement of
Granular Computing has attracted some researchers to
study. LIANG et al. (2015) believe that multi-granular
would be used as a new approach to data mining, and listed
some multi-granular applications in big data. Yao (2016)
suggests using a multi-granular structure to achieve a
multi-level understanding at varying granularity layers and
to get a multi-view from different angles. We can compare
or combine different granules to gain additional insights
that are not available from a sing le view. Multi-granular is
& Shu Zhao
zhaoshuzs2002@hotmail.com
1
Key Laboratory of Intelligent Computing and Signal
Processing, Ministry of Education, Anhui University,
Hefei 230601, Anhui Province, China
2
School of Computer Science and Technology, Anhui
University, Hefei 230601, Anhui, People’s Republic of China
123
Granular Computing
https://doi.org/10.1007/s41066-019-00194-2
(0123456789().,-volV)(0123456789().,-volV)