User Tools

Site Tools


algorithms:communities_ev

Leading Eigenvector Communities

Detects communities using a divisive process based on the leading eigenvector of the modularity matrix. A community is a set of nodes with more edges inside than with other communities. We use the power method to calculate that eigenvector for each network division, maximizing the modularity of the network partition. As in many community detection algorithms, this method does not guarantee the optimal solution of the problem, only an approximate solution. Moreover, it has some clear limitations in particular networks. Read the documentation for more information.

Input parameter Type Default Description
Iterations* int > 0 100 Number of iterations in the power method to calculate the eigenvector.
Resolution* float > 0 1.0 Resolution parameter for the modularity, which controls the number of communities identified. Reduce it to increase the number of communities, and increase it to reduce the number of communities.
Algorithm iterations int > 0 1 Number of time that the whole algorithm is run with different random initialization. The final output is the partition that maximizes the modularity.
Link Weight text None Link property name that plays the role of strength. The property values should be positive numbers.

* Required Field


The following two plots show the partition in communities of the Zachary's Karate Club network using resolution parameters 1.0 and 0.5, giving two differetn partitions.



References:

  • M. Newman, Networks: An introduction, Oxford University Press, 2010, ISBN: 978-0-19-920665-0, sec. 11.8, page 375.
  • M. Newman, Modularity and community structure in networks, PNAS Applied Mathematics vol. 103 no. 23 – 8577-8582 (2006).
algorithms/communities_ev.txt · Last modified: 2018/12/20 14:22 by systems