Clustering, Visualizing, and Navigating for Large Dynamic Graphs

Abstract

In this paper, we present a new approach to exploring dynamic graphs. We have developed a new clustering algorithm for dynamic graphs which finds an ideal clustering for each time-step and links the clusters together. The resulting time-varying clusters are then used to define two visual representations. The first view is an overview that shows how clusters evolve over time and provides an interface to find and select interesting time-steps. The second view consists of a node link diagram of a selected time-step which uses the clustering to efficiently define the layout. By using the time-dependant clustering, we ensure the stability of our visualization and preserve user mental map by minimizing node motion, while simultaneously producing an ideal layout for each time step. Also, as the clustering is computed ahead of time, the second view updates in linear time which allows for interactivity even for graphs with upwards of tens of thousands of nodes.

Temporal Multivariate Networks

“…Sallaberry et al [94] cluster every timestep individually, associate the clusters across time, and use the space-filling curve approach to render each timestep; see Fig. 4.…”

“…With large networks, stability becomes even more important, but so does "motion coherency". Even small motions on each vertex are too much to perceive if they are chaotic, but if vertices move coherently, they can be perceived as a single group [94]. In a series of papers Archambault and Purchase evaluate various approaches for dynamic graph visualization and consider how they affect mental map preservation [7,4,6], also summarized in a recent survey [5].…”

“…However, producing good results with these algorithms is computationally expensive, as they do not scale well to large data sets. To apply storyline techniques to dynamic graphs, an intermediary step of dynamic clustering must be derived [88,94].…”

“…Hu et al [58] use a similar approach, but apply a heuristic to accelerate this process. Sallaberry et al [94], on the other hand, cluster each timestep separately and then use Jaccard index to track the clusters across time.…”

Hierarchical Spatio-temporal Visual Analysis of Cluster Evolution in Electrocorticography Data

“…The arbitrary ordering of the nodes in the vertical axis may increase link crossings between axes, inhibiting easy comprehension of the evolution patterns. To address this issue, Reda et al and Sallabury et al [27,31] employ sorting techniques to place active and stable communities at the top of the vertical axis.…”