Analysis of Large Two-mode Networks using Pajek

Analysis of Large Two-mode Networks using Pajek

Instructor: Vladimir Batagelj (University of Ljubljana, Slovenia)

 

The data from data sources (data bases, web services, WWW) can be transformed into (often large) two-mode networks. For example: (people, movies, took part in creation of), (buyers/consumers, goods, quantity), (parliamentarians, problems, positive vote), (people, events, took part in), etc.

The usual approach to analysis of two-mode networks is by projection to a one-mode network based on its first or second set on which the traditional network analysis methods are applied. We show that this transformation can be done in different ways (using different normalizations) leading to substantially different results. We also present some direct methods for two-mode network analysis: degree distributions, variant of Kleinberg's hubs and authorities, two-mode cores, and 4-rings weights.

In some cases we can get from data sources collections of compatible (having the same first set) two mode networks. For example from bibliographies (BibTeX, Web of Science, Scopus, CiteSeer, Zentralblatt MATH, Google Scholar, and others) we can obtain two-mode networks: works X authors (WA), works X keywords (WK), works X journal/publisher (WJ), and the one-mode citation network (Ci) between works. From these networks additional networks can be derived using network multiplication. For example (AW is the transpose of WA):
- (standard) collaboration network = AW * WA,
- authors using keywords = AW * WK,
- citations between authors = AW * Ci * WA,
- author co-citations = AW * CiT * Ci * WA,
(CiT is the transpose of Ci) and others.

In the workshop the participants will learn how to analyze the obtained networks using clustering, islands and other methods for discovering important subnetworks. The obtained subnetworks are often very dense and the usual "dots and lines" visualization doesn't give readable results. We demonstrate alternative approaches such as matrix representation (with ordering based on clustering or blockmodeling) and drawing of skeleton (Pathfinder).

The workshop is based on the free (for non-commercial use) program Pajek. The 64-bit version of Pajek enables the user to analyze networks with some tens of millions of nodes.

References:

1. Batagelj, V., Cerinšek, M.: On bibliographic networks. Scientometrics, 96(2013)3, 845-864
2. De Nooy, W., Mrvar, A., Batagelj, V.: Exploratory Social Network Analysis with Pajek; Revised and Expanded Second Edition. Structural Analysis in the Social Sciences, Cambridge University Press, September 2011.
3. Kejžar, N., Korenjak Černe, S., Batagelj, V.: Network Analysis of Works on Clustering and Classification from Web of Science. Classification as a Tool for Research. Hermann Locarek-Junge, Claus Weihs eds. Proceedings of IFCS 2009. Studies in Classification, Data Analysis, and Knowledge Organization, 525-536, Springer, Berlin, 2010.
4. Pajek’s wiki. http://pajek.imfm.si

 

 

Campus d'excel·lència internacional U A B