Figure 3

Network degree distribution. We show the edge distribution for two networks, one built using the top 5,000 pairs according to the unweighted hypergeometric metric without runs and one from our runs-employing two-term model. We see that the unweighted hypergeometric network contains many more edges of high degree. In particular, nodes with more than 40 edges are almost completely absent from the runs network while being abundant in the unweighted hypergeometric network. This suggests that the runs-informed network contains smaller and more interpretable clusters.