creating network data from event level data - r

Question

As a newbie to network analysis, I am struggling with transforming an event level dataset I want to plot into the correct shape. I am grateful for any hints/ leads/ etc. What I did until now, broadly follows this introduction.
The dataset in question contains events organized by the political party Jobbik. Each event defined by a unique id (id) has associated organizational sponsors (org_names) and their type (org). There is no hierarchy between org_1, org_2, or org_names1 and org_names2.
Originally the dataset comes in a wide format. Although I am not sure if this is what I should be doing, the first step I do is to transform the data into a long format and clean a bit the strings. This is the code for reading in the data and getting it into a long format:
jobbik <- read.csv("http://eborbath.github.io/stackoverflow/jobbik.csv")
library(tidyverse)
library(stringr)
library(igraph)
# long format
jobbik <- reshape(as.data.frame(jobbik), dir='long',
varying=list(c(3:13), c(14:24)),
v.names=c('org_names', 'org'), times = c(as.character(seq(1:11))))
jobbik$org <- str_trim(jobbik$org, side="both")
jobbik$org_names <- str_trim(jobbik$org_names, side="both")
jobbik <- jobbik %>%
filter(!(org=="no other organizer" & org_names=="")) %>%
filter(!(org=="JOBBIK" & org_names %in% c("Jobbik",
"Jobbik Magyarországért Mozgalom",
"",
"JObbik",
"jobbik",
"aktivisté Jobbiku",
"a Jobbik"))) %>%
mutate(org_names=ifelse(org_names=="", org, org_names)) %>%
distinct(.)
In the next step I want to create the network dataset. To do so, I calculate the number of times each unique organization has been involved in events with Jobbik. Add Jobbik as one side of each edge and plot the data with igraph:
network <- jobbik %>%
select(id, org_names) %>%
group_by(org_names) %>%
summarise(weight = n()) %>%
ungroup() %>%
mutate(from=1,
org_names=as.factor(org_names)) %>%
mutate(org_id=as.numeric(factor(org_names)))
edges <- network %>% select(from, org_id, weight)
nodes <- network %>% select(org_id, org_names) %>%
mutate(org_names=as.character(org_names))
routes_igraph <- graph_from_data_frame(d = edges, vertices = nodes, directed = FALSE)
plot(routes_igraph, layout = layout_with_graphopt)
While this runs and creates the network, it only gets me the relationship between each unique organization and Jobbik, but not the relationship between these organizations, which do not involve Jobbik. I realize that the error is in the data transformation I do and I should use the event level information to calculate the number of times each organizational pair has been involved in organizing something together, then plot that data. Unfortunately, though I don't know how to get there. I am grateful for any help.

Answer 1

I am not exactly an expert in network analysis, and igraph in particular. But I think something like that might be helpful.
I changed the preprocessing part of your analysis, because I've found few complications in a way:
Encoding of Hungarian language: that took time to find right encoding (see locale = 'cp1250 in read_csv call;
After gathering I've changed org_name* to org and org* into type;
I use chop to make it easier to spread -> unnest;
I've tried to make filter call shorter, but with no big success;
I use stringr::str_to_title() to unify org var, because there are same names which differs only in the way that nth word of the name is capitalized or not;
I use coalesce to fill NAs of org var with values from type var.
library(tidyverse)
library(magrittr)
library(igraph)
jobbik <- read_csv(
"http://eborbath.github.io/stackoverflow/jobbik.csv",
trim_ws = T,
locale = locale(encoding = 'cp1250')
)
jobbik %<>%
gather('key', 'val', -c('id', 'date')) %>%
mutate(
key = case_when(
grepl('^org_names\\d+$', key) ~ 'org',
grepl('^org\\d+$', key) ~ 'type',
TRUE ~ key
)
) %>%
chop(val) %>%
spread(key, val) %>%
unnest(c(org, type)) %>%
filter(
!(is.na(org) & (type == 'no other organizer')) &
!((is.na(org) | grepl('.*jobbik.*', org, T )) & (type == 'JOBBIK'))
) %>%
mutate(org = str_to_title(coalesce(org, type)))
To form data frame of graph edges, I am grouping by id of the event, filtering out all the events that where supported by only one organization (so there is no connection with other organizations), and finally I create pairs within id between the organizations with combn function. The result is character vector Org A-Org B, which, after unnesting, I separate into to cols from and to using - as a split (which is potentially dangerous, if the name of the org. has - symbol in it). I also filter out all self loops, if any. The last operation is count, to calculate how frequently each individual pair appears through the list of Jobbik meetings. I assign it to the width because when plotting, igraph::plot will use it as a width for the edges.
ed <- jobbik %>%
group_by(id) %>%
filter(n() > 1) %>%
summarise(edge = list(combn(org, 2, paste, collapse = '-'))) %>%
unnest(edge) %>%
separate(edge, into = c('from', 'to'), sep = '-') %>%
filter(from != to) %>%
count(from, to, name = width)
Similar analysis is performed for vertices. I add here extra information for the vertices, namely event id, date, organization type which you could use further, color - mapping the number of times given org. supported Jobbik and some additional graphical parameters for latter plot.
nd <- jobbik %>%
filter(org %in% c(ed$from, ed$to)) %>%
group_by(name = org) %>%
summarise(
id = sprintf('Event ids: %s', paste(id, collapse = ', ')),
date = sprintf('Event dates: %s', paste(date, collapse = ', ')),
type = sprintf('Org. type: %s', paste(type, collapse = '; ')),
color = n()
) %>%
ungroup() %>%
mutate(
color = heat.colors(10)[cut(color, 10)],
frame.color = NA,
label.dist = 1,
label.cex = .5,
label.color = 'gray10'
)
With these data we can make undirected graph, using graph_from_data_frame() function:
g <- graph_from_data_frame(ed, F, nd)
vertex_attr(g, 'size') <- degree(g, mode = 'all')
In a second line above, I add vertex attribute size to map degree of the vertices to the size of the vertices.
And finally to plot the comunity, I can do just:
plot(
g,
edge.curved = .2,
layout = layout_with_kk,
asp = 1,
main = 'Jobbik interaction network',
)