Novel (?) technique for inferring a directed bipartite projection from an undirected bipartite graph. Code is for pedagogical demonstration to accompany the article here: http://dmarx.github.io/map-of-reddit-by-active-users/

undirected_to_directed_bipartite_projection.R

library(igraph)

# Experiment parameters
n=10 # Primary class (i.e. subreddits)
m=100 # Secondary class (i.e. users)
threshold = .5 # edge threshold

######################################

seed(123)
g = sample_bipartite(n,m, p=.5) # this will generate a random graph, rather than scale free.
plot(g)
mat = as_adj(g)
#dim(mat) # NB: "(n+m) x (n+m)" form (block off-diagonal adjacency), rather than n x m (bipartite adjacency)
mat_bipart_sp = mat[1:n, 1:m +n] # Extract n x m form
mat_bipart = as.matrix(mat_bipart_sp) # sparse matrix strangely doesn't support transpose operation

# calculate node degree
n_deg = apply(mat_bipart_sp, 1, sum)

# Perform projection operation relative to first class
bipart_proj = as.matrix(mat_bipart_sp %*% t(mat_bipart))

# Recalibrate edge weights according to target degree
dir_bipat = t(t(as.matrix(bipart_proj)) / n_deg) # need to take transpose to weight by target degree
diag(dir_bipat) = 0

dir_bipat_trsh = dir_bipat
dir_bipat_trsh[dir_bipat<threshold] = 0

# visualize
g2 = graph_from_adjacency_matrix(dir_bipat_trsh, weighted=TRUE)
plot(g2)