Figure 2

Scaling of average degree for each class of node reveals sparse and dense layers. Scaling of the average degree \(\langle k_{X} \rangle \) with the number of documents \(n_{D}\) depends on the node types X. The average degree was computed over all nodes of the same type (see legend, where H,T,M indicates the layer) in a sample of \(n_{D}\) documents from dataset. The symbols (error bars) are the average (standard deviation) over multiple random samples of documents. The prediction for the degree of word types using Eq. (1) is also plotted for reference