Social network plot

[CKres]

I have a confidential version of my data in which I would like to make a social network using igraph. :

It is a "node list" that is paired with an "edge list" to make a social network using igraph. I however have been difficulty making my data appear weighted, ( I would assume that the individuals within my data set with the highest weights would be centered in a plot, but they have not been in any version I've made. Is there an easy fix to ensure that the igraph plot actually appears weighted instead of random?

[nirgrahamuk]

Are you aware of igraph layouts ?

Also, what is the meaning of a weight 0 in your context? If it implies no connectivity then it may be sensible to filter it out of the data altogether ?

[CKres]

I am not very familiar with igraph layouts , yes you are correct, filtering all null values would be sensible. Do you have any suggestions as to what igraph layout to use?

[nirgrahamuk]

Here is an example with fruchterman reingold

library(igraph) 
library(tidyverse)
set.seed(42)
spread_factor <- 20
node_size_factor <-2000

(edgelist <- data.frame(from=sample(letters,100,replace=TRUE),
                       to=sample(letters,100,replace=TRUE)) %>% filter(
                         from!=to
                       ) %>% arrange(from,to))

wgt_df <- enframe(sort(table(unlist(edgelist))^3),
        value="weight")

edgelist2 <- left_join(edgelist,
                       wgt_df,by=c("from"="name")) %>%  left_join(wgt_df,
                                                                  by=c("to"="name")) %>%
  mutate(weight=(weight.x*weight.y)/sum(weight.x*weight.y)/spread_factor)

g <- graph_from_data_frame(edgelist2)
is_weighted(g)

enframe(sort(strength(g))) %>% 
  mutate(name=forcats::as_factor(name)) %>% 
  ggplot(aes(x=as_factor(as.integer(name)),y=value)) + geom_text(aes(label=name))

plot(g,layout= layout_with_fr(g),rescale=TRUE,vertex.size=strength(g)*node_size_factor)

Q and E are the most connected.

[CKres]

@nirgrahamuk

edgelist2 <- left_join(edgelist,
                       wgt_df,by=c("from"="name")) %>%  left_join(wgt_df,
                                                                  by=c("to"="name")) %>%
  mutate(weight=(weight.x*weight.y)/sum(weight.x*weight.y)/spread_factor

Hi nirgrahamuk,

I tried to follow some of the similar code you provided with the example.
I've attached my code below and a figure from it.
There are two issues I keep encountering:

1. Each node moves location in the plot every time the plot line is rerun
2. The individuals with the lowest number (1:~10) I hypothesized would be central due to their weights being higher. Is there a modification that can be made in the plot to display this?
   
   

library(readr)

Import Data

link <- read_csv("~/Desktop/sting.edge.csv")
node <- read.csv("~/Desktop/sting.node.csv")

examine

head(node)
head(link)
nrow(node); length(unique(node$Name))
nrow(link); nrow(unique(link[,c("From", "To")]))

link <- link[order(link$From, link$To),]
colnames(link)[3] <- "weight"
rownames(link) <- NULL

library('igraph')

Weights

wgt_df2 <- enframe(sort(table(unlist(link))^3),
value="weight")

edgelist2b <- left_join(link,
wgt_df2,by=c("From"="name")) %>% left_join(wgt_df2,
by=c("To"="name")) %>%
mutate(weight=(weight.xweight.y)/sum(weight.xweight.y)/spread_factor)

plotting

g2 <- graph_from_data_frame(link)
is_weighted(g2)
??node_size_factor
enframe(sort(strength(g2))) %>%
mutate(name=forcats::as_factor(name)) %>%
ggplot(aes(x=as_factor(as.integer(name)),y=value)) + geom_text(aes(label=name))
node_size_factor <-.03
plot(g2,layout= layout_with_fr(g2),rescale=TRUE,vertex.size=strength(g2)*node_size_factor,
remove.multiple=F, remove.loops=T, edge.arrow.size=.001, vertex.color="lightblue")

[nirgrahamuk]

The vertex size is scaled to the strength of the connections yet yours all seem pretty much the same to my eye, if 1 to 10 are stronger it cant be by very much ?

[CKres]

Yes they are, however, the vertices on my plot do not stay in the same location every time I run the same plot. Is this normal?

[nirgrahamuk]

Yes because layout_with_fr() uses a iterative algorithm with a random initialization you should use set.seed() for reproducibilty

[CKres]

Ah, that makes sense. Thank you nirgrahamuk, I appreciate all of your wonderful insight!