[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [igraph] community detection

From: Bernie Hogan
Subject: Re: [igraph] community detection
Date: Tue, 14 Aug 2012 15:27:29 -0400

The resolution parameter in the Louvain method might make a difference. I'm pretty sure that the igraph (multilevel_community) implementation uses only a resolution parameter of 1, but this is tunable in other implementations. 

Check out the examples on this page:

Even if igraph doesn't implement these methods, it does play very nice with pajek for import and export. 

The other alternative is to consider overlapping clustering routines, such as OSLOM (which at the moment is only implemented in C and is a bit tedious to work with). 

Take care,

Dr Bernie Hogan
Research Fellow, Oxford Internet Institute
University of Oxford

On 14 Aug 2012, at 12:05, Stijn van Dongen <address@hidden> wrote:

Your graph seems quite dense; about (2*3M) / 7K = 857 outgoing
edges/arcs per node on average.  This is *only* on average.

If you have, in addition, a graph that has highly connected hub nodess
and/or small diameter, it could well be a very hard graph to cluster.

I advise you to look into ways to reduce the number of edges in the
graph and make sure it does not have highly connected nodes; a simple
approach is possible if the edges have weights associated with them.  (I
recommend considering (reciprocal) nearest neighbour type selection;
e.g. take the arc merge (or intersect) of all top-N sections of the arc
lists of all nodes).

It should be possible even in the absence of weights, either with quick
and crude methods (remove nodes with many edges) or by considering the
number of triangles an edge participates in and thus establishing some
kind of weighting.

There is another clustering algorithm that has a parameter affecting
granularity (mcl - I wrote it), but I have not yet made it available
in igraph unfortunately. Anyway, I think the considerations above are
likely more important.


On Tue, Aug 14, 2012 at 11:31:31AM -0400, Sam Steingold wrote:
No matter what community detection I use (leading.eigenvector.community,
multilevel.community &c), I get just 2 communities (I have 7k vertexes
and ~3M edges).  I want many more smaller communities (even if less
modular), because this community detection is just one step in a long
How do I limit the community size?
Is there a better way to ensure that I get many communities?
Sam Steingold (http://sds.podval.org/) on Ubuntu 12.04 (precise) X 11.0.11103000
http://www.childpsy.net/ http://dhimmi.com http://mideasttruth.com
http://americancensorship.org http://palestinefacts.org http://camera.org
will write code that writes code that writes code for food

igraph-help mailing list

Stijn van Dongen         >8<        -o)   O<  forename pronunciation: [Stan]
EMBL-EBI                            /\\   Tel: +44-(0)1223-492675
Hinxton, Cambridge, CB10 1SD, UK   _\_/   http://micans.org/stijn

igraph-help mailing list

reply via email to

[Prev in Thread] Current Thread [Next in Thread]