Seeing Networks Change

Network visualization helps us understand local and global connectivity patterns of single networks. Force-directed layouts are most common and incorporate heuristic strategies and aesthetic criteria to emphasize symmetry and... [ view full abstract ]

Network visualization helps us understand local and global connectivity patterns of single networks. Force-directed layouts are most common and incorporate heuristic strategies and aesthetic criteria to emphasize symmetry and improve the relative layout of neighboring nodes. Unfortunately, they are not helpful when networks need to be compared—even minor differences in small networks cannot be dependably detected. To address this deficiency, we introduce differential hive plots.

Hive plots (www.hiveplot.com) are a network visualization strategy that place nodes and edges using an absolute coordinate system along radially arranged axes. Rules that define the coordinate system are flexible, making these plots tunable to be sensitive to overall network patterns. Hive plots are deterministic and perceptually uniform–small changes in the network appear as small changes in its hive plot–making it possible to compare them visually. However, subtle differences in hive plots of large networks can be difficult to detect, particularly in regions where node and edge density is high.

To assist in using hive plots for pair-wise comparison of networks, we use the hive plot of each network to derive a differential hive plot. This plot is generated by performing a set operation on two hive plots using tunable difference criteria and cutoffs. For example, the intersection of two hive plots shows the similarities in two networks, whereas a symmetric set difference operation highlights the differences.