Table 1.
Number of non-singleton communities and singletons by community type: S(tructural), A(ctivity-based), T(opic-based), and I(nteraction-based).
Fig 1.
The proportion of communities greater than s in size, across the different community types.
Note the logarithmic scale on the horizontal and vertical axes.
Fig 2.
The normalized mutual information between the coverings inferred from the different community types.
Values of normalized mutual information close to 1 indicate similarity in the community structure, while values close to 0 indicate dissimilarity. The normalized mutual information is computed with singletons and orphan nodes included. Note the block-diagonal structure, indicating the strong relationship between question type and community membership.
Fig 3.
A schematic of how, given a covering, the edges of the network can be partitioned using (8) into inter-edges, intra-edges, and mixed-edges.
Inter-edges (red dashed) cross community boundaries. Intra-edges (blue solid) remain inside community boundaries. Mixed-edges (purple dotted) both remain in and cross community boundaries due to overlap in community membership. Each column corresponds to the same collection of weights, but partitioned using a different covering. Each row corresponds to the same covering, but for the different weights.
Fig 4.
The density of edge weights for different community types (rows) and weight types (columns).
The red, blue, and purple values below each collection of densities indicate the median of weight on intra-, inter-, and mixed-edges, respectively.