Fast random walk with restart and its applications

Hanghang Tong, Christos Faloutsos, Jia Yu Pan

Research output: Chapter in Book/Report/Conference proceedingConference contribution

519 Citations (Scopus)

Abstract

How closely related are two nodes in a graph? How to compute this score quickly, on huge, disk-resident, real graphs? Random walk with restart (RWR) provides a good relevance score between two nodes in a weighted graph, and it has been successfully used in numerous settings, like automatic captioning of images, generalizations to the "connection subgraphs", personalized PageRank, and many more. However, the straightforward implementations of RWR do not scale for large graphs, requiring either quadratic space and cubic pre-computation time, or slow response time on queries. We propose fast solutions to this problem. The heart of our approach is to exploit two important properties shared by many real graphs: (a) linear correlations and (b) blockwise, community-like structure. We exploit the linearity by using low-rank matrix approximation, and the community structure by graph partitioning, followed by the Sherman-Morrison lemma for matrix inversion. Experimental results on the Corel image and the DBLP dabasets demonstrate that our proposed methods achieve significant savings over the straightforward implementations: they can save several orders of magnitude in pre-computation and storage cost, and they achieve up to 150x speed up with 90%+ quality preservation.

Original languageEnglish (US)
Title of host publicationProceedings - IEEE International Conference on Data Mining, ICDM
Pages613-622
Number of pages10
DOIs
StatePublished - 2006
Externally publishedYes
Event6th International Conference on Data Mining, ICDM 2006 - Hong Kong, China
Duration: Dec 18 2006Dec 22 2006

Other

Other6th International Conference on Data Mining, ICDM 2006
CountryChina
CityHong Kong
Period12/18/0612/22/06

Fingerprint

Costs

ASJC Scopus subject areas

  • Engineering(all)

Cite this

Tong, H., Faloutsos, C., & Pan, J. Y. (2006). Fast random walk with restart and its applications. In Proceedings - IEEE International Conference on Data Mining, ICDM (pp. 613-622). [4053087] https://doi.org/10.1109/ICDM.2006.70

Fast random walk with restart and its applications. / Tong, Hanghang; Faloutsos, Christos; Pan, Jia Yu.

Proceedings - IEEE International Conference on Data Mining, ICDM. 2006. p. 613-622 4053087.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Tong, H, Faloutsos, C & Pan, JY 2006, Fast random walk with restart and its applications. in Proceedings - IEEE International Conference on Data Mining, ICDM., 4053087, pp. 613-622, 6th International Conference on Data Mining, ICDM 2006, Hong Kong, China, 12/18/06. https://doi.org/10.1109/ICDM.2006.70
Tong H, Faloutsos C, Pan JY. Fast random walk with restart and its applications. In Proceedings - IEEE International Conference on Data Mining, ICDM. 2006. p. 613-622. 4053087 https://doi.org/10.1109/ICDM.2006.70
Tong, Hanghang ; Faloutsos, Christos ; Pan, Jia Yu. / Fast random walk with restart and its applications. Proceedings - IEEE International Conference on Data Mining, ICDM. 2006. pp. 613-622
@inproceedings{e2a620c137074d66b6c22f12778e61ed,
title = "Fast random walk with restart and its applications",
abstract = "How closely related are two nodes in a graph? How to compute this score quickly, on huge, disk-resident, real graphs? Random walk with restart (RWR) provides a good relevance score between two nodes in a weighted graph, and it has been successfully used in numerous settings, like automatic captioning of images, generalizations to the {"}connection subgraphs{"}, personalized PageRank, and many more. However, the straightforward implementations of RWR do not scale for large graphs, requiring either quadratic space and cubic pre-computation time, or slow response time on queries. We propose fast solutions to this problem. The heart of our approach is to exploit two important properties shared by many real graphs: (a) linear correlations and (b) blockwise, community-like structure. We exploit the linearity by using low-rank matrix approximation, and the community structure by graph partitioning, followed by the Sherman-Morrison lemma for matrix inversion. Experimental results on the Corel image and the DBLP dabasets demonstrate that our proposed methods achieve significant savings over the straightforward implementations: they can save several orders of magnitude in pre-computation and storage cost, and they achieve up to 150x speed up with 90{\%}+ quality preservation.",
author = "Hanghang Tong and Christos Faloutsos and Pan, {Jia Yu}",
year = "2006",
doi = "10.1109/ICDM.2006.70",
language = "English (US)",
isbn = "0769527019",
pages = "613--622",
booktitle = "Proceedings - IEEE International Conference on Data Mining, ICDM",

}

TY - GEN

T1 - Fast random walk with restart and its applications

AU - Tong, Hanghang

AU - Faloutsos, Christos

AU - Pan, Jia Yu

PY - 2006

Y1 - 2006

N2 - How closely related are two nodes in a graph? How to compute this score quickly, on huge, disk-resident, real graphs? Random walk with restart (RWR) provides a good relevance score between two nodes in a weighted graph, and it has been successfully used in numerous settings, like automatic captioning of images, generalizations to the "connection subgraphs", personalized PageRank, and many more. However, the straightforward implementations of RWR do not scale for large graphs, requiring either quadratic space and cubic pre-computation time, or slow response time on queries. We propose fast solutions to this problem. The heart of our approach is to exploit two important properties shared by many real graphs: (a) linear correlations and (b) blockwise, community-like structure. We exploit the linearity by using low-rank matrix approximation, and the community structure by graph partitioning, followed by the Sherman-Morrison lemma for matrix inversion. Experimental results on the Corel image and the DBLP dabasets demonstrate that our proposed methods achieve significant savings over the straightforward implementations: they can save several orders of magnitude in pre-computation and storage cost, and they achieve up to 150x speed up with 90%+ quality preservation.

AB - How closely related are two nodes in a graph? How to compute this score quickly, on huge, disk-resident, real graphs? Random walk with restart (RWR) provides a good relevance score between two nodes in a weighted graph, and it has been successfully used in numerous settings, like automatic captioning of images, generalizations to the "connection subgraphs", personalized PageRank, and many more. However, the straightforward implementations of RWR do not scale for large graphs, requiring either quadratic space and cubic pre-computation time, or slow response time on queries. We propose fast solutions to this problem. The heart of our approach is to exploit two important properties shared by many real graphs: (a) linear correlations and (b) blockwise, community-like structure. We exploit the linearity by using low-rank matrix approximation, and the community structure by graph partitioning, followed by the Sherman-Morrison lemma for matrix inversion. Experimental results on the Corel image and the DBLP dabasets demonstrate that our proposed methods achieve significant savings over the straightforward implementations: they can save several orders of magnitude in pre-computation and storage cost, and they achieve up to 150x speed up with 90%+ quality preservation.

UR - http://www.scopus.com/inward/record.url?scp=34748827346&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=34748827346&partnerID=8YFLogxK

U2 - 10.1109/ICDM.2006.70

DO - 10.1109/ICDM.2006.70

M3 - Conference contribution

AN - SCOPUS:34748827346

SN - 0769527019

SN - 9780769527017

SP - 613

EP - 622

BT - Proceedings - IEEE International Conference on Data Mining, ICDM

ER -