Ndtyjky

I have a dataframe of 380 observations of 9 variables. The data represents the cooperation between persons doing similar projects. In the first column is the main node and the other columns represent the persons s/he cooperated with on a project which each column representing one person. So if perchance researcher in row 1 column 1 cooperated with five persons, their names will be in five columns, and if researcher in row 2 column 1 cooperated with 3 persons their names will be in the other three columns. Obviously there will be many empty columns as not all researchers cooperate with same amount of persons. With this data, how do I plot this into a network graph?

dataframe example:

data <- data.frame(

author_1 = c('John', 'Kerry', 'Michelle', 'Joan', 'Paul'),

author_2 = c('Joan', 'Rick', 'N/A', 'Terrence', 'Collin'),

author_3 = c('Terrence', 'Michelle', 'Michelle', 'Joan', 'Paul'),

author_4 = c('Michelle', 'Collin', 'N/A', 'N/A', 'Phillips'))

I tried using graph.data.frame but that only gives connection between the first two columns.

We can try with the ggraph package, but we have to arrange well the data.

# this are your data

data <- data.frame(

author_1 = c('John', 'Kerry', 'Michelle', 'Joan', 'Paul'),

author_2 = c('Joan', 'Rick', 'N/A', 'Terrence', 'Collin'),

author_3 = c('Terrence', 'Michelle', 'Michelle', 'Joan', 'Paul'),

author_4 = c('Michelle', 'Collin', 'N/A', 'N/A', 'Phillips'))



# here you load some nice package

library(tidyr)      # to tidy the data

library(ggraph)     # to plot nice network data with the semantic of ggplot

library(tidygraph)  # to work with networks

library(ggrepel)    # to not have overlapping labels

First, you should prepare your data. Due you have a father row, author_1, and sons, you can manage to do this for each combinations of author_1 and author_n, due you should have only one columns. It clearly works also if you have not a hierarchical dataset. You should have all the combinations of doubles father-sons for each rows, and rbind() do it, merging all the combinations (easier to do than to explain).

edges <-rbind(

expand(data, nesting(author_1,author_2))  %>%  `colnames<-`(c("a", "b")),  # for 1 and 2, we do all the combinations and give name a and b

expand(data, nesting(author_1,author_3))  %>%  `colnames<-`(c("a", "b")),  # for 1 and 3, we do all the combinations and give name a and b

expand(data, nesting(author_1,author_4))  %>%  `colnames<-`(c("a", "b"))   # for 1 and 3, we do all the combinations and give name a and b

)

edges

# A tibble: 15 x 2

   a        b       

   <fct>    <fct>   

 1 Joan     Terrence

 2 John     Joan    

 3 Kerry    Rick    

 4 Michelle N/A     

 5 Paul     Collin  

 6 Joan     Joan    

 7 John     Terrence

 8 Kerry    Michelle

 9 Michelle Michelle

10 Paul     Paul    

11 Joan     N/A     

12 John     Michelle

13 Kerry    Collin  

14 Michelle N/A     

15 Paul     Phillips

Remember, if you would plot the N/A, you left this as is, in other hand you add at the end this %>% filter(b != 'N/A') .

Now we manage the data to put them in the graph:

# create edges

edges1 <- edges%>% group_by(a,b) %>% summarise(weight = sum(n()))



# create nodes

nodes <- rbind(data.frame(researcher = edges$a, n = 1),data.frame(researcher = edges$b, n = 1)) %>%

  group_by(researcher) %>%

  summarise(n = sum(n))



# now we have to have the match between edges and nodes

edges1$a <- match(edges1$a, nodes$researcher) 

edges1$b <- match(edges1$b, nodes$researcher)



# declare the data as graph data

tidy <- tbl_graph(nodes = nodes, edges = edges1, directed = T)

tidy <- tidy %>% 

  activate(edges) %>% 

  arrange(desc(weight)

  ) 



# now the plot: you have several options to do, here a basic one

ggraph(tidy, layout = "gem") +     

geom_node_point(aes(size=n)) +                          # size of the node the frequency

geom_edge_link(aes(width = weight),                     # here you set the edges

                                                        # thickness as frequency

               arrow = arrow(length = unit(4, 'mm')),   # arrows, if you want

               end_cap = circle(3, 'mm'), alpha = 0.8) + 

scale_edge_width(range = c(0.2, 2)) +

geom_text_repel(aes(x = x, y=y , label=researcher)) 

That should be coherent with data and:

> edges1

# A tibble: 14 x 3

# Groups:   a [?]

       a     b weight

   <int> <int>  <int>

 1     1     1      1

 2     1     7      1

 3     1     9      1

 4     2     1      1

 5     2     9      1

 6     2     4      1

 7     3     6      1

 8     3     8      1

 9     3     4      1

10     4     7      2

11     4     4      1

12     5     6      1

13     5     5      1

14     5    10      1

> nodes

# A tibble: 10 x 2

   researcher     n

   <fct>      <dbl>

 1 Joan           5

 2 John           3

 3 Kerry          3

 4 Michelle       6

 5 Paul           4

 6 Collin         2

 7 N/A            3

 8 Rick           1

 9 Terrence       2

10 Phillips       1