49 Network Visualization in R

Yunze Pan

49.1 Introduction

visNetwork is a powerful tool in R to help us describe networks and explore the structure visually. It is extremely useful for us to obtain valuable information from an interactive network graph. In this tutorial, we will offer a quick introduction for newcomers to learn concepts of creating networks in R. Hope you will enjoy!

49.2 Installation

The main packages we are going to use for network visualization in R are visNetwork and igraph. They can be installed with install.packages(“visNetwork”) and install.packages(“igraph”).

49.3 Dataframe

In this section we will create a small network that simulates student interactions on campus. Our objective is to get you familiar with using visNetwork as quickly as as possible. In order to visualize interactive networks, we will first read two datasets (a nodes data.frame and an edges data.frame). Then, we can explore the various layout options by adding different variables on our nodes data.frame and edges data.frame.

49.3.1 Nodes

A nodes data.frame must include a id column. Each id represents the node we want to display in our graph. Other optional columns can also be added into our nodes data.frame. They can help us to distinguish nodes in our graph. For example, each node is a student with a unique assigned id, his/her name, major, and major.type.

nodes <- data.frame(id=1:7, # id column (must be called id)
                    name=c("Asher","Bella","Chloe","Daniel","Emma","Frank","Gabriel"), # student names
                    major=c("CS","CS","CS","STAT","DS","DS","DS"), # CS: computer science major, STAT: statistics major, DS: data science major
                    major.type=c(1,1,1,2,3,3,3)) # 1: CS, 2: STAT, 3: DS
##   id    name major major.type
## 1  1   Asher    CS          1
## 2  2   Bella    CS          1
## 3  3   Chloe    CS          1
## 4  4  Daniel  STAT          2
## 5  5    Emma    DS          3
## 6  6   Frank    DS          3
## 7  7 Gabriel    DS          3

49.3.2 Edges

An edges data.frame must include a from column and a to column denoting the starting node and ending node of each edge. We use id to represent the starting node and ending node. We also add a weight column on our edges data.frame to describe the frequency of interactions between two nodes. For example, in the first row, we know student 1 reached out to student 2 once.

edges <- data.frame(from=c(1,1,2,3,5,5,6,7),
##   from to weight
## 1    1  2      1
## 2    1  4      1
## 3    2  3      1
## 4    3  1      1
## 5    5  4      1
## 6    5  6      1
## 7    6  7      1
## 8    7  5      1

49.4 Visualiztion

Now we can visualize our student interaction network using visNetwork. Examples are showed as below. We will start from the default setting and then move on to customize our network for a better interactive visualization.

49.4.1 Minimal Example

visNetwork(nodes, edges)

49.4.2 Customize Node

colors <- colorRampPalette(brewer.pal(3, "RdBu"))(3) # use three colors to distinguish students by their majors
nodes <- nodes %>% mutate(shape="dot", # "shape" variable: customize shape of nodes ("dot", "square", "triangle")
                          shadow=TRUE, # "shadow" variable: include/exclude shadow of nodes
                          title=major, # "title" variable: tooltip (html or character), when the mouse is above
                          label=name, # "label" variable: add labels on nodes
                          size=20, # "size" variable: set size of nodes
                          borderWidth=1, # "borderWidth" variable: set border width of nodes
                          color.background=colors[major.type], # "color.background" variable: set color of nodes
                          color.border="grey", # "color.border" variable: set frame color
                          color.highlight.background="yellow", # "color.highlight.background" variable: set color of the selected node
                          color.highlight.border="black") # "color.highlight.border" variable: set frame color of the selected node
visNetwork(nodes, edges, width="100%", main="Student Interaction Network") %>% # "main" variable: add a title
  visLayout(randomSeed=4) # give a random seed manually so that the layout will be the same every time

49.4.3 Customize Edge

edges <- edges %>% mutate(width=weight*3, # "width" variable: set width of each edge
                          color="lightgrey", # "color" variable: set color of edges
                          arrows="to", # "arrows" variable: set arrow for each edge ("to", "middle", "from ")
                          smooth=TRUE) # "smooth" variable: each edge to be curved or not
visNetwork(nodes, edges, width="100%", main="Student Interaction Network") %>% 

49.4.4 Add Legend Based on Groups

nodes <- nodes %>% mutate(group=major) # add a "group" column on node data.frame and add groups on nodes
visNetwork(nodes, edges, width="100%", main="Student Interaction Network") %>%
  visLayout(randomSeed=4) %>% 
  visGroups(groupname="CS", color=colors[1]) %>% # color "colors[1]" for "CS" group 
  visGroups(groupname="STAT", color=colors[2]) %>%
  visGroups(groupname="DS", color=colors[3]) %>%
  visLegend(width=0.1, position="right", main="Academic Major") # "position" variable: set position ("left", "right") 

49.4.5 Select by Node

nodes <- nodes %>% select(-group) # remove "group" column because we don't want to show legend this time
visNetwork(nodes, edges, width="100%", main="Student Interaction Network") %>%
  visLayout(randomSeed=4) %>% 
  visOptions(nodesIdSelection=TRUE, # "nodesIdSelection" variable: select a node by id
             selectedBy="major") %>% # "selectedBy" variable: select a node by the values of a column such as "major" column

49.4.6 Highlight Nearest Nodes

visNetwork(nodes, edges, width="100%", main="Student Interaction Network") %>% 
  visLayout(randomSeed=4) %>% 
  visOptions(highlightNearest = list(enabled = TRUE, # "enabled" variable: highlight nearest nodes and edges by clicking on a node
                                     degree = 2)) # "degree" variable: set degree of depth

49.4.7 Edit Network

visNetwork(nodes, edges, width="100%", main="Student Interaction Network") %>%
  visLayout(randomSeed=4) %>% 
  visOptions(highlightNearest=TRUE, # degree of depth = 1
             manipulation=TRUE) %>%  # "manipulation" variable: add/delete nodes/edges or change edges

49.4.8 Add Navigation Buttons and Control Interactions

visNetwork(nodes, edges, width="100%", main="Student Interaction Network") %>%
  visLayout(randomSeed=4) %>%
             selectedBy="major") %>% 
  visInteraction(hideEdgesOnDrag=TRUE, # "hideEdgesOnDrag" variable: hide edges when dragging the view
                 dragNodes=TRUE, # "dragNodes" variable: hide nodes when dragging the view
                 dragView=TRUE, # "dragView" variable: enable or not the movement of the full network
                 zoomView=TRUE, # "zoomView" variable: enable or not the zoom (use mouse scroll)
                 navigationButtons=TRUE) %>% # "navigationButtons" variable: show navigation buttons

49.5 Export

Finally, we use visSave() to save network in html file.

our_network <- visNetwork(nodes, edges)
visSave(our_network, file = "Student Interaction Network.html", background="white")

49.6 Help?

More information about visNetwork.


49.7 Social Network Analysis

We have already learned how to visualize the interactive network. To help you better understand its application, we will use visNetwork and igraph to perform our social network analysis.

49.7.1 Dataset

We will investigate interactions in the movie Star Wars Episode IV. First, we import two csv files (“nodes.csv” and “edges.csv”). Each node in “nodes.csv” is a character and each edge in “edges.csv” tells whether two characters appeared together in a scene of the movie. Thus, edges are undirected. Since characters may appear in multiple scenes together, each edge has a weight.

sw_nodes <- read.csv("https://raw.githubusercontent.com/pablobarbera/data-science-workshop/master/sna/data/star-wars-network-nodes.csv")
##          name id
## 1       R2-D2  0
## 2   CHEWBACCA  1
## 3       C-3PO  2
## 4        LUKE  3
## 6       CAMIE  5
sw_edges <- read.csv("https://raw.githubusercontent.com/pablobarbera/data-science-workshop/master/sna/data/star-wars-network-edges.csv")
##      source target weight
## 1     C-3PO  R2-D2     17
## 2      LUKE  R2-D2     13
## 3   OBI-WAN  R2-D2      6
## 4      LEIA  R2-D2      5
## 5       HAN  R2-D2      5
## 6 CHEWBACCA  R2-D2      3

We group our characters (“dark side” or “light side” or “other”).

dark_side <- c("DARTH VADER", "MOTTI", "TARKIN")
other <- c("GREEDO", "JABBA")
sw_nodes$group <- NA
sw_nodes$group[sw_nodes$name %in% dark_side] <- "dark side"
sw_nodes$group[sw_nodes$name %in% light_side] <- "light side"
sw_nodes$group[sw_nodes$name %in% other] <- "other"

Let’s try another network package called igraph to explore the network.

First, we use the graph_from_data_frame function, which needs two arguments: d and vertices. The igraph object g indicates that there are 22 nodes and 66 edges.

g <- graph_from_data_frame(d=sw_edges, vertices=sw_nodes, directed=FALSE) # an undirected graph
## IGRAPH 8dd7562 UNW- 22 60 -- 
## + attr: name (v/c), id (v/n), group (v/c), weight (e/n)
## + edges from 8dd7562 (vertex names):
##  [1] R2-D2      --C-3PO       R2-D2      --LUKE        R2-D2      --OBI-WAN    
##  [4] R2-D2      --LEIA        R2-D2      --HAN         R2-D2      --CHEWBACCA  
##  [7] R2-D2      --DODONNA     CHEWBACCA  --OBI-WAN     CHEWBACCA  --C-3PO      
## [10] CHEWBACCA  --LUKE        CHEWBACCA  --HAN         CHEWBACCA  --LEIA       
## [16] CAMIE      --BIGGS       LUKE       --BIGGS       DARTH VADER--LEIA       
## [19] LUKE       --BERU        BERU       --OWEN        C-3PO      --BERU       
## [22] LUKE       --OWEN        C-3PO      --LUKE        C-3PO      --OWEN       
## + ... omitted several edges

Next, we output a portion of the adjacency matrix for our network.

g[1:6, 1:6] # the first six rows and columns
## 6 x 6 sparse Matrix of class "dgCMatrix"
## R2-D2           .         3    17   13           .     .
## CHEWBACCA       3         .     5   16           1     .
## C-3PO          17         5     .   18           .     .
## LUKE           13        16    18    .           .     2
## DARTH VADER     .         1     .    .           .     .
## CAMIE           .         .     .    2           .     .

49.7.2 Visualization

Alternatively, we can show a heat map of our adjacency matrix. The number in each square equals to the weight of one edge. We observe LUKE is a very popular character.

sw_matrix <- as.matrix(g[])
sw_matrix <- sw_matrix[order(rownames(sw_matrix)), order(colnames(sw_matrix))]
melted_sw_matrix <- melt(sw_matrix)
ggplot(melted_sw_matrix, aes(x=Var1, y=Var2, fill=value)) + 
  geom_tile() +
  geom_text(aes(label=value), color="red") +
  scale_fill_gradient(low="white", high="black") +
  xlab("characters") + ylab("characters") +
  theme(axis.text.x=element_text(angle=45)) +

We also compute characters’ importance using strength() function based on the number of scenes they appear in and rank the importance in a descending order. The goal of strength() function is to sum up the edge weights of the adjacent edges for each node.

importance <- strength(g)
sw_nodes$importance <- importance
head(arrange(sw_nodes, -importance))
##        name id      group importance
## 1      LUKE  3 light side        129
## 2       HAN 13 light side         80
## 3     C-3PO  2 light side         64
## 4 CHEWBACCA  1 light side         63
## 5      LEIA  7 light side         59
## 6     R2-D2  0 light side         50

Again, we use visNetwork to visualize.

sw_colors <- colorRampPalette(brewer.pal(3, "RdBu"))(3)
sw_nodes$group.type <- NA
sw_nodes$group.type[sw_nodes$group=="dark side"] <- sw_colors[1]
sw_nodes$group.type[sw_nodes$group=="other"] <- sw_colors[2]
sw_nodes$group.type[sw_nodes$group=="light side"] <- sw_colors[3]
sw_nodes <- sw_nodes %>% select(-id) %>%
         size=log((importance+3)^5), # adjust size with respect to a node's importance
         color.highlight.border="black") %>% 
sw_edges <- sw_edges %>% mutate(from=source,
                                width=log((weight+3)^1.5), # adjust width with respect to an edge's weight
visNetwork(sw_nodes, sw_edges, width="100%", main="Star Wars Episode IV Network") %>%
  visLayout(randomSeed=21) %>% 
  visGroups(groupname="dark side", color=sw_colors[1]) %>%
  visGroups(groupname="other", color=sw_colors[2]) %>%
  visGroups(groupname="light side", color=sw_colors[3]) %>%
  visLegend(width=0.1, position="right", main="Group") %>%
             selectedBy="group") %>% 

You may wonder how important a character is in our Star Wars network. Therefore, we want to utilize three proposed measures (degree centrality, betweenness centrality, and closeness centrality) to quantify each node’s importance in a network and visualize how its importance is different from others.

49.7.3 Centrality Measurement

Degree centrality is defined as the number of adjacent edges to each node. After ranking the degree centrality, we find LUKE has the greatest value. It implies that LUKE is interacting with a great amount of unique characters. We color each node based on its degree centrality value. The node with the greatest value has the warmest color.

degree_centrality <- degree(g)
sw_nodes$degree_centrality <- degree_centrality[as.character(sw_nodes$name)]
head(sort(degree_centrality, decreasing=TRUE))
##      LUKE      LEIA     C-3PO CHEWBACCA       HAN     R2-D2 
##        15        12        10         8         8         7
sw_colors_centrality <- rev(colorRampPalette(brewer.pal(9, "Oranges"))(22))
sw_nodes <- sw_nodes %>% mutate(degree_rank=23-floor(rank(degree_centrality)),
network_degree <- visNetwork(sw_nodes, sw_edges, height='350px', width="100%", main="Degree Centrality") %>%
  visLayout(randomSeed=21) %>% 
             selectedBy="degree_rank") %>% 

Betweenness centrality is defined as the number of shortest paths between nodes that pass through a particular node. After ranking the betweenness centrality, we find LEIA has the greatest value. It implies that LEIA tends to be very critical to the communication process. We color each node based on its betweenness centrality value. The node with the greatest value has the warmest color.

betweenness_centrality <- betweenness(g)
sw_nodes$betweenness_centrality <- betweenness_centrality[as.character(sw_nodes$name)]
head(sort(betweenness_centrality, decreasing=TRUE))
##       LEIA    DODONNA        HAN      C-3PO      BIGGS RED LEADER 
##   59.95000   47.53333   37.00000   32.78333   31.91667   31.41667

Closeness centrality is defined as the number of steps required to access every other node from a given node. After ranking the closeness centrality, we find BIGGS has the greatest value. It implies that BIGGS is close to many other characters. We color each node based on its closeness centrality value. The node with the greatest value has the warmest color.

closeness_centrality <- closeness(g, normalized=TRUE)
sw_nodes$closeness_centrality <- closeness_centrality[as.character(sw_nodes$name)]
head(sort(closeness_centrality, decreasing=TRUE))
##   0.3448276   0.3448276   0.3333333   0.3333333   0.3278689   0.3125000

Lastly, we output our network and find discrepancies among three measurements.

49.8 External Resource

  1. visNetwork package;
  2. star-wars-network.