Network Visualization II

Social Network Analysis

Termeh Shafie

From last time

# load and manipulate data
data("starwars")
sw1 <- starwars[[1]]
sw_palette <- c("#1A5878", "#C44237", "#AD8941", "#E99093", "#50594B")
V(sw1)$interactions <- graph.strength(sw1) 

#plot
ggraph(graph = sw1,layout = "stress") + 
  geom_edge_link0(edge_colour = "grey25",
                  aes(edge_linewidth = weight)) +
  geom_node_point(shape = 21, color = "black",stroke = 1,
                  aes(fill = sex,size = interactions)) +
  geom_node_text(color = "black", size = 4, repel = FALSE, 
                 aes(filter = (interactions>=65),label = name))+
  scale_edge_width(range = c(0.1,1.5),guide = "none")+
  scale_size(range = c(3,10),guide = "none")+
  scale_fill_manual(values = sw_palette, na.value = "grey",name = "")+
  coord_fixed()+
  theme_graph() +
  theme(legend.position = "bottom") +
  guides(fill = guide_legend(override.aes = list(size=6)))

From last time

Go beyond the standard layout

layout stress is sufficient for most network visualization tasks

Ego centric layout

Emphasize the position of certain nodes in the network.

ggraph(sw1,layout = "focus",focus = 19)+
  draw_circle(col = "#00BFFF", use = "focus",max.circle = 3)+
  geom_edge_link0(edge_colour = "grey25",edge_alpha = 0.5)+
  geom_node_point(shape = 21,size = 5,fill = "grey66")+
  geom_node_text(aes(filter = (name=="ANAKIN"),label = name))+
  theme_graph()+
  coord_fixed()

• focus=... : id to be put in the center (other nodes are on concentric circles around it)
  
• draw_circle: draw the concentric circles

Ego centric layout

Centrality layout

concentric circle layout according to a centrality index

strength <- graph.strength(sw1)
ggraph(sw1,layout = "centrality",cent = strength)+
  draw_circle(col = "#00BFFF", use = "cent")+
  geom_edge_link0(edge_colour = "grey25",edge_alpha = 0.5)+
  geom_node_point(shape = 21,size = 5,fill = "grey66")+
  geom_node_text(aes(filter = (strength>=45),label = name),repel = TRUE)+
  theme_graph()+
  coord_fixed()

Centrality layout

Backbone layout

layout_as_backbone() can help emphasize hidden group structures

g <- sample_islands(9,40,0.4,15)
g <- simplify(g)
V(g)$grp <- as.character(rep(1:9,each = 40))

try the standard first

ggraph(g,layout = "stress")+
  geom_edge_link0(edge_colour = "black",edge_linewidth = 0.1, edge_alpha = 0.5)+
  geom_node_point(shape = 21, size = 3, aes(fill = grp))+
  scale_fill_brewer(palette = "Set1")+
  theme_graph()+
  theme(legend.position = "none")

Backbone layout

Backbone layout

try to reveal the hidden group structure with layout="backbone"

ggraph(g, layout = "backbone")+
  geom_edge_link0(edge_colour = "black",edge_linewidth = 0.1, edge_alpha = 0.5)+
  geom_node_point(shape = 21,size = 3, aes(fill = grp))+
  scale_fill_brewer(palette = "Set1")+
  theme_graph()+
  theme(legend.position = "none")

Backbone layout

Backbone layout

Facebook friendships of a university.
Node colour corresponds to dormitory of students

Large networks

layout="sparse_stress" and layout="pmds" work well for up to 50k nodes

graph	stress	sparsestress(50)	sparsestress(100)	drl(igraph)	mds(igraph)	pivotmds(50)	pivotmds(100)
1138bus (n=1138, m=1458)	2.6s	0.4s	0.7s	1.7s	1.4s	0.05s	0.08s
3elt (n=4720, m=13722)	113s	3.4s	5.2s	8.4s	95s	1s	1.1s
power grid (n=4941, m=6594)	134s	3.4s	5.6s	6.9s	114s	1s	1.1s
pesa (n=11738 m=67828)	-	18.3s	26.5s	21s	-	6.3s	6.8s
cond-mat(n=13861 m=44619)	-	22.9s	33.5s	24.8s	-	8.0s	8.7s

Layout summary

most useful layouts
layout = "stress": all purpose layout algorithm
layout = "focus": ego-centric type layouts
layout = "centrality": concentric centrality layout
layout = "backbone": emphasize a group structure (if it exists)
layout = "sparse_stress": large networks

Miscellaneous

Do not recompute layout continuously

lay <- create_layout(g,"stress")

ggraph(lay) + 
  geom_edge_link0() +
  geom_node_point()

But need to recompute the layout if attributes change!

lay <- layout_with_stress(g)

ggraph(g, "manual", x = xy[,1],y = xy[,2]) + 
  geom_edge_link0() +
  geom_node_point()

Longitudinal networks

A series of snapshots visualized with individual nodes easy to trace

Example: 50 actor excerpt from the Teenage Friends and Lifestyle Study

layout_as_dynamic() is used to visualize with fixed node positions

The implemented algorithm calculates a reference layout which is a layout of the union of all networks and individual layouts based on stress minimization and combines those in a linear combination which is controlled by the alpha parameter. For alpha=1, only the reference layout is used and all graphs have the same layout. For alpha=0, the stress layout of each individual graph is used. Values in-between interpolate between the two layouts.

xy <- layout_as_dynamic(s50, alpha = 0.2)

Use ggraph in conjunction with patchwork to produce static plots

Longitudinal networks

pList <- vector("list", length(s50))

for (i in 1:length(s50)) {
    pList[[i]] <- ggraph(s50[[i]], layout = "manual", x = xy[[i]][, 1], y = xy[[i]][, 2]) +
        geom_edge_link0(edge_linewidth = 0.6, edge_colour = "grey66") +
        geom_node_point(shape = 21, aes(fill = as.factor(smoke)), size = 6) +
        geom_node_text(label = 1:50, repel = FALSE, color = "white", size = 4) +
        scale_fill_manual(
            values = c("forestgreen", "grey25", "firebrick"),
            guide = ifelse(i != 2, "none", "legend"),
            name = "smoking",
            labels = c("never", "occasionally", "regularly")
        ) +
        theme_graph() +
        theme(legend.position = "bottom") +
        labs(title = paste0("Wave ", i))
}

wrap_plots(pList)

Longitudinal networks

Can we do better?

yes, we can animate the changes…

Longitudinal networks

Use the ggforce

The ggforce package works pretty nicely with ggraph

library(ggforce)

Use the ggforce

ggraph(g, #|
    layout = "manual",
    x = bb$xy[, 1],
    y = bb$xy[, 2]
) +
    geom_edge_link0(aes(col = col), width = 0.2) +
    geom_node_point(aes(fill = grp), shape = 21, size = 3) +
    geom_mark_hull(
        aes(x, y, group = grp, fill = grp),
        concavity = 4,
        expand = unit(2, "mm"),
        alpha = 0.25
    ) +
    scale_color_brewer(palette = "Set1") +
    scale_fill_brewer(palette = "Set1") +
    scale_edge_color_manual(values = c(rgb(0, 0, 0, 0.3), rgb(0, 0, 0, 1))) +
    theme_graph() +
    theme(legend.position = "none")

Use the ggforce

you can also add labels to your cluster

Use the ggforce

you can also add labels to your cluster

ggraph(g, #|
    layout = "manual",
    x = bb$xy[, 1],
    y = bb$xy[, 2]
) +
    geom_edge_link0(aes(col = col), width = 0.2) +
    geom_node_point(aes(fill = grp), shape = 21, size = 3) +
    geom_mark_hull(
        aes(x, y, group = grp, fill = grp, label = grp),
        concavity = 4,
        expand = unit(2, "mm"),
        alpha = 0.25
    ) +
    scale_color_brewer(palette = "Set1") +
    scale_fill_brewer(palette = "Set1") +
    scale_edge_color_manual(values = c(rgb(0, 0, 0, 0.3), rgb(0, 0, 0, 1))) +
    theme_graph() +
    theme(legend.position = "none")

Use the ggforce

want to avoid node overlaps?
use geom_node_voronoi() orginiated from geom_voronoi_tile()

Small tricks for common problems

How can I achieve that my directed edges stop at the node border, independent from the node size?

Small tricks for common problems

How can I achieve that my directed edges stop at the node border, independent from the node size?

Small tricks for common problems

The overlap can be avoided by using the I() function from base R, which treats the entries of a vector “as is”. So we know that if a node has degree 5, it will be mapped to a circle with radius (or diameter?) “5pt”. Since this means, that you have no control over the scaling, you need to do that beforehand.

## this function is borrowed from the ambient package
normalise <- function(x, from = range(x), to = c(0, 1)) {
    x <- (x - from[1]) / (from[2] - from[1])
    if (!identical(to, c(0, 1))) {
        x <- x * (to[2] - to[1]) + to[1]
    }
    x
}

## map to the range you want
V(g)$degree <- normalise(V(g)$degree, to = c(3, 11))

ggraph(g, "stress") +
    geom_edge_link(
        aes(end_cap = circle(node2.degree + 2, "pt")),
        edge_colour = "grey25",
        arrow = arrow(
            angle = 10,
            length = unit(0.15, "inches"),
            ends = "last",
            type = "closed"
        )
    ) +
    geom_node_point(aes(size = I(degree)), col = "grey66") +
    theme_graph()

Small tricks for common problems

How can I lower opacity of nodes without making edges visible underneath?

Small tricks for common problems

How can I lower opacity of nodes without making edges visible underneath?

The solution is rather simple. Just add a node layer with the same aesthetics below with alpha=1 (default) and color=“white” (or the background color of the plot).

Small tricks for common problems

ggraph(g, "stress") +
    geom_edge_link(edge_colour = "grey66") +
    geom_node_point(size = 10, col = "white") +
    geom_node_point(
        aes(alpha = degree),
        size = 10,
        col = "red",
        show.legend = FALSE
    ) +
    theme_graph()

Small tricks for common problems

How can I enhance readability of node labels in hairball graphs?

Small tricks for common problems

How can I enhance readability of node labels in hairball graphs?

Here you can make use of the fact that the layout of the nodes are stored in a “hidden” data frame when a ggraph object is constructed. That means you can use other geoms from other packages. In this case, the shadowtext package as shown below.

Small tricks for common problems

ggraph(g, "stress") +
    geom_edge_link0(aes(edge_color = weight, edge_linewidth = weight), show.legend = FALSE) +
    geom_node_point(size = 8, color = "#44a6c6") +
    shadowtext::geom_shadowtext(aes(x, y, label = name), color = "black", size = 4, bg.colour = "white") +
    scale_edge_color_continuous(low = "grey66", high = "black") +
    scale_edge_width(range = c(0.1, 0.5)) +
    theme_graph() +
    coord_fixed()

Edgebundling & flow maps

Using the package edgebundle