***** To join INSNA, visit http://www.insna.org *****
Mason,

Thank you so much. Both links point to important material that I was unaware of. As I replied to James Moody, all I meant by lumpier was that in this case, the Louvain method produces a few clusters that are double or triple the size of the clusters produced by the VOS method. If I were to chart the two distributions, the Louvain distribution would have a few larger peaks. It looks like the variance is greater to me. I am not a mathematician or statistician, but I am curious about why these differences occur. I am aware that different methods are likely to produce different results, but am interested in this particular case. 

Thanks again.

John


On Mon, Aug 5, 2013 at 4:42 PM, Mason Alexander Porter <[log in to unmask]> wrote:
*****  To join INSNA, visit http://www.insna.org  *****

John,

I'm not sure what you mean by 'lumpier', but let me just make the generic comment that one _expects_ different methods in which a complicated optimization landscape is involved to give different results.  There is an extensive discussion in the following paper (in the context of modularity optimization, including with Louvain):

        http://pre.aps.org/abstract/PRE/v81/i4/e046106



I have never played with VOS, but in my googling to figure out what that is, I found this web page (which seems concerned with rather similar issues as what you raise):

        http://mrvar.fdv.uni-lj.si/pajek/community/LouvainVOS.htm



------
Mason





On Mon, 5 Aug 2013, SOCNET automatic digest system wrote:

There is 1 message totalling 363 lines in this issue.

Topics of the day:

 1. Louvain vs VOS Community Detection

_____________________________________________________________________
SOCNET is a service of INSNA, the professional association for social
network researchers (http://www.insna.org). To unsubscribe, send
an email message to [log in to unmask] containing the line
UNSUBSCRIBE SOCNET in the body of the message.

----------------------------------------------------------------------

Date:    Mon, 5 Aug 2013 10:58:35 +0900
From:    John McCreery <[log in to unmask]>
Subject: Louvain vs VOS Community Detection

--047d7b5d6592708e4304e329a524
Content-Type: text/plain; charset=ISO-8859-1

*****  To join INSNA, visit http://www.insna.org  *****

Two recent runs through part of my data using the Pajek commands for,
first, VOS and, second, Louvain community detection. My question is why the
results using the Louvain method seem lumpier than those using the VOS
method. Note, for example, that cluster 1 in the VOS distribution has 36
members, while cluster 1 using the Louvain method has 97. Using the VOS
method, cluster 11 has 35 members, while cluster 11 using the Louvain
method has 101. Is there a straightforward explanation why the two
algorithms produce such apparently different results? Or is this just a
random event?

Thoughts?



-----
Mason

----------------------------------------------------------------------------
 Mason A. Porter
 University Lecturer
 Oxford Centre for Industrial and Applied Mathematics
 Mathematical Institute, University of Oxford

 Homepage: http://people.maths.ox.ac.uk/porterm/
 Blog: http://masonporter.blogspot.com/
 Twitter: @masonporter
 Skype: tepid451
 ----------------------------------------------------------------------------
 "Few things are deadlier than me armed with a red pen." (Me)
 ----------------------------------------------------------------------------

_____________________________________________________________________
SOCNET is a service of INSNA, the professional association for social
network researchers (http://www.insna.org). To unsubscribe, send
an email message to [log in to unmask] containing the line
UNSUBSCRIBE SOCNET in the body of the message.



--
John McCreery
The Word Works, Ltd., Yokohama, JAPAN
Tel. +81-45-314-9324
[log in to unmask]
http://www.wordworks.jp/
_____________________________________________________________________ SOCNET is a service of INSNA, the professional association for social network researchers (http://www.insna.org). To unsubscribe, send an email message to [log in to unmask] containing the line UNSUBSCRIBE SOCNET in the body of the message.