Comparative network analysis using KronFit

Gupta Sukrit, Puzis Rami, Kilimnik Konstantin

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

1 Scopus citations

Abstract

Comparative network analysis is an emerging line of research that provides insights into the structure and dynamics of networks by finding similarities and discrepancies in their topologies. Unfortunately, comparing networks directly is not feasible on large scales. Existing works resort to representing networks with vectors of features extracted from their topologies and employ various distance metrics to compare between these feature vectors. In this paper, instead of relying on feature vectors to represent the studied networks, we suggest fitting a network model (such as Kronecker Graph) to encode the network structure. We present the directed fitting-distance measure, where the distance from a network A to another network B is captured by the quality of B’s fit to the model derived from A. Evaluation on five classes of real networks shows that KronFit based distances perform surprisingly well.

Original languageEnglish
Title of host publicationComplex Networks VII - Proceedings of the 7th Workshop on Complex Networks CompleNet 2016
EditorsHocine Cherifi, Bruno Goncalves, Ronaldo Menezes, Roberta Sinatra
PublisherSpringer Verlag
Pages363-375
Number of pages13
ISBN (Print)9783319305684
DOIs
StatePublished - 1 Jan 2016
Event7th Workshop on Complex Networks CompleNet, 2016 - Dijon, France
Duration: 23 Mar 201625 Mar 2016

Publication series

NameStudies in Computational Intelligence
Volume644
ISSN (Print)1860-949X

Conference

Conference7th Workshop on Complex Networks CompleNet, 2016
Country/TerritoryFrance
CityDijon
Period23/03/1625/03/16

Keywords

  • Comparative analysis
  • Complex networks
  • Distance metrics
  • Generative models

ASJC Scopus subject areas

  • Artificial Intelligence

Fingerprint

Dive into the research topics of 'Comparative network analysis using KronFit'. Together they form a unique fingerprint.

Cite this