RoClust: Role discovery for graph clustering

Bin Hui Chou, Einoshin Suzuki

Research output: Contribution to journalArticle

Abstract

Graph clustering, or community detection, is an important task of discovering the underlying structure in a network by clustering vertices in a graph into communities. In the past decades, non-overlapping methods such as normalized cuts and modularity-based methods, which assume that each vertex belongs to a single community, are proposed to discover disjoint communities. On the other hand, overlapping methods such as CPM, which assume that each vertex can belong to multiple communities, have been drawing increasing attention as the assumption fits the reality. In this paper, we show that existing non-overlapping and overlapping methods lack consideration to edges that link a vertex to its neighbors belonging to different communities, which often leads to counter-intuitive results of vertices located near borders of communities. Therefore, we propose a new graph clustering methods named RoClust, which uses three roles, bridges, gateways and hubs to discover communities. Each of the three roles represents a kind of vertices that connect communities. Experimental results show that RoClust outperforms state-of-the-art methods of graph clustering including non-overlapping and overlapping methods.

Original languageEnglish
Pages (from-to)1-20
Number of pages20
JournalWeb Intelligence and Agent Systems
Volume11
Issue number1
DOIs
Publication statusPublished - Jul 22 2013

All Science Journal Classification (ASJC) codes

  • Software
  • Computer Networks and Communications
  • Artificial Intelligence

Cite this

RoClust : Role discovery for graph clustering. / Chou, Bin Hui; Suzuki, Einoshin.

In: Web Intelligence and Agent Systems, Vol. 11, No. 1, 22.07.2013, p. 1-20.

Research output: Contribution to journalArticle

@article{ff6aaf397a9147eda2904f6f57dc3109,
title = "RoClust: Role discovery for graph clustering",
abstract = "Graph clustering, or community detection, is an important task of discovering the underlying structure in a network by clustering vertices in a graph into communities. In the past decades, non-overlapping methods such as normalized cuts and modularity-based methods, which assume that each vertex belongs to a single community, are proposed to discover disjoint communities. On the other hand, overlapping methods such as CPM, which assume that each vertex can belong to multiple communities, have been drawing increasing attention as the assumption fits the reality. In this paper, we show that existing non-overlapping and overlapping methods lack consideration to edges that link a vertex to its neighbors belonging to different communities, which often leads to counter-intuitive results of vertices located near borders of communities. Therefore, we propose a new graph clustering methods named RoClust, which uses three roles, bridges, gateways and hubs to discover communities. Each of the three roles represents a kind of vertices that connect communities. Experimental results show that RoClust outperforms state-of-the-art methods of graph clustering including non-overlapping and overlapping methods.",
author = "Chou, {Bin Hui} and Einoshin Suzuki",
year = "2013",
month = "7",
day = "22",
doi = "10.3233/WIA-130259",
language = "English",
volume = "11",
pages = "1--20",
journal = "Web Intelligence",
issn = "2405-6456",
publisher = "IOS Press",
number = "1",

}

TY - JOUR

T1 - RoClust

T2 - Role discovery for graph clustering

AU - Chou, Bin Hui

AU - Suzuki, Einoshin

PY - 2013/7/22

Y1 - 2013/7/22

N2 - Graph clustering, or community detection, is an important task of discovering the underlying structure in a network by clustering vertices in a graph into communities. In the past decades, non-overlapping methods such as normalized cuts and modularity-based methods, which assume that each vertex belongs to a single community, are proposed to discover disjoint communities. On the other hand, overlapping methods such as CPM, which assume that each vertex can belong to multiple communities, have been drawing increasing attention as the assumption fits the reality. In this paper, we show that existing non-overlapping and overlapping methods lack consideration to edges that link a vertex to its neighbors belonging to different communities, which often leads to counter-intuitive results of vertices located near borders of communities. Therefore, we propose a new graph clustering methods named RoClust, which uses three roles, bridges, gateways and hubs to discover communities. Each of the three roles represents a kind of vertices that connect communities. Experimental results show that RoClust outperforms state-of-the-art methods of graph clustering including non-overlapping and overlapping methods.

AB - Graph clustering, or community detection, is an important task of discovering the underlying structure in a network by clustering vertices in a graph into communities. In the past decades, non-overlapping methods such as normalized cuts and modularity-based methods, which assume that each vertex belongs to a single community, are proposed to discover disjoint communities. On the other hand, overlapping methods such as CPM, which assume that each vertex can belong to multiple communities, have been drawing increasing attention as the assumption fits the reality. In this paper, we show that existing non-overlapping and overlapping methods lack consideration to edges that link a vertex to its neighbors belonging to different communities, which often leads to counter-intuitive results of vertices located near borders of communities. Therefore, we propose a new graph clustering methods named RoClust, which uses three roles, bridges, gateways and hubs to discover communities. Each of the three roles represents a kind of vertices that connect communities. Experimental results show that RoClust outperforms state-of-the-art methods of graph clustering including non-overlapping and overlapping methods.

UR - http://www.scopus.com/inward/record.url?scp=84880215499&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84880215499&partnerID=8YFLogxK

U2 - 10.3233/WIA-130259

DO - 10.3233/WIA-130259

M3 - Article

AN - SCOPUS:84880215499

VL - 11

SP - 1

EP - 20

JO - Web Intelligence

JF - Web Intelligence

SN - 2405-6456

IS - 1

ER -