net_edge_betweenness() returns a copy of a network in which each edge's
weight is replaced by its betweenness: the number of shortest paths
between ordered pairs of distinct nodes that pass through that edge, with
a fractional share when several shortest paths tie (Girvan & Newman
2002). It
takes a single network or a netobject_group, which is scored one
network at a time.
With invert = TRUE (the default) a heavier edge is a shorter step, so
the shortest path between two states of a transition network is the most
probable route, not the one with the fewest hops.
Only positive weights count as edges, and a self-loop never lies on a
shortest path, so its betweenness is 0. A directed network gives an
asymmetric betweenness matrix; an undirected one a symmetric matrix,
each cell holding the betweenness of that edge. The result keeps the
source network's directedness (directed when it records none).
An edge with high betweenness carries many of the shortest routes
between states; Girvan & Newman (2002) read such edges as the bridges
between groups of states. To compare the edge betweenness of
two groups, pass two results to permutation() or bayes_compare(),
which resample the source networks behind them; both need two results
with the same source method and invert, and bayes_compare() needs
"relative" source networks.
Usage
net_edge_betweenness(x, invert = TRUE, ...)
# S3 method for class 'netobject'
net_edge_betweenness(x, invert = TRUE, ...)
# S3 method for class 'netobject_group'
net_edge_betweenness(x, invert = TRUE, ...)
# Default S3 method
net_edge_betweenness(x, invert = TRUE, ...)
# S3 method for class 'net_edge_betweenness'
plot(x, style = c("bar", "forest", "delta"), top_n = NULL, labels = TRUE, ...)Arguments
- x
A
netobjectornetobject_group; any other input raises an error. For theplot()method: anet_edge_betweennessnetwork.- invert
Whether to turn weights into distances by
1 / weightbefore finding shortest paths, which suits probability and frequency networks. WithFALSE, the weight itself is the distance; use it when the weights already are distances. DefaultTRUE.- ...
Not used.
- style
For the
plot()method: the kind of chart."bar"(default) draws one horizontal bar per edge;"forest"a stem from zero to a point, with a dashed line at the mean betweenness of the edges shown;"delta"each edge's deviation from that mean as a bar pointing right (above the mean, blue) or left (below, red).- top_n
For the
plot()method: the number of highest edges to keep. DefaultNULL(every edge with non-zero betweenness).- labels
For the
plot()method: whether to print each edge's value (its deviation, for"delta") beside it. DefaultTRUE.
Value
For a netobject: a network of class
c("net_edge_betweenness", "netobject", "cograph_network") with method
"edge_betweenness", whose weights are the edge betweenness scores.
It prints like any network; extract_edges() gives one row per edge,
plot() ranks the edges and cograph::splot() draws the network. The
source network's data, metadata, parameters, weights and initial
probabilities are kept on it, which is what lets permutation() and
bayes_compare() test differences in edge betweenness between two
such networks. For a netobject_group: a netobject_group of these
networks, one per group.
For the plot() method: a ggplot object.
Details
Shortest paths come from all-pairs Floyd-Warshall on the distance
matrix (1 / weight, or the weight with invert = FALSE), counting tied
paths with a relative tolerance so that floating-point rounding does not
break a tie. The values agree with tna::betweenness_network() for
transition networks, and on an undirected network each cell equals
igraph::edge_betweenness() for that edge; the function is named
differently to avoid masking either.
Methods
plot() ranks the edges with non-zero betweenness, highest first, one
row per edge (from -> to). It needs no cograph; cograph::splot()
draws the same object as a node-link diagram. A network whose edges all
have zero betweenness raises an error.
References
Girvan, M. & Newman, M. E. J. (2002). Community structure in social and biological networks. Proceedings of the National Academy of Sciences, 99(12), 7821–7826. doi:10.1073/pnas.122653799
Examples
seqs <- data.frame(
V1 = c("A","B","A","C"), V2 = c("B","C","B","A"),
V3 = c("C","A","C","B"))
net <- build_network(seqs, method = "relative")
eb <- net_edge_betweenness(net)
extract_edges(eb)
#> from to weight
#> 1 C A 3
#> 2 A B 3
#> 3 B C 3