[igraph] multilevel community in weighted networks

[Stefano Breschi]

I am applying community detection algorithms to a weighted network. The weight is just a number comprised between 0 and 1. What is puzzling me is that the partition returned by the multilevel.community method seems to depend on the scaling of the weights, i.e. if one multiplies the weight by 10 etc., whereas the fastgreedy.community method is not affected by this scaling issue, i.e. the partitions are always the same irrespective of the scale.

Here is the link to the data I am using

https://www.dropbox.com/s/yxw7pjrsgqbfjsj/example.csv

The code applied is the following:

library(igraph)

a <- read.csv("example.csv",header=TRUE)

h <- graph.data.frame(a, directed=FALSE)

h <- simplify(h, remove.multiple=TRUE, remove.loops=TRUE)

memberships <- list()

fg <- fastgreedy.community(h,weights=E(h)$weight)

memberships$`fg` <- fg$membership

fg10 <- fastgreedy.community(h,weights=E(h)$weight*10)

memberships$`fg10` <- fg10$membership

fg100 <- fastgreedy.community(h,weights=E(h)$weight*100)

memberships$`fg100` <- fg100$membership

lm <- multilevel.community(h,weights=E(h)$weight) 

memberships$`lm` <- lm$membership

lm10 <- multilevel.community(h,weights=E(h)$weight*10) 

memberships$`lm10` <- lm10$membership

lm100 <- multilevel.community(h,weights=E(h)$weight*100) 

memberships$`lm100` <- lm100$membership

names=V(h)$name

out <- as.data.frame(memberships)

commun <- cbind(names,out)

Any help would be appreciated. Thanks a lot for the wonderful package you maintain.

Stefano 

--

Department of Management & Technology and KITes

Università L. Bocconi
Via Roentgen 3
20136 Milan (Italy)
Grafton Building 5th floor
Office B1-02
tel. +39-02-5836 3365
fax  +39-02-5836 3399