Topology-aware Visualization for Interactive Graph Structure Exploration

Structural analysis of graphs can reveal significant connectivity patterns in areas of sociology, biology, etc. However, discovering specific pattern types is complex and time consuming. Besides, users can hardly recognize them via visual clues. In this paper, we propose a supervised dual-task model to predict the topological category of the graph, i.e., whether it approximates egocentric, grid, and clique patterns. Meanwhile, the model generates drawings that reflect the characteristics of these patterns to enhance user cognition. It is robust to local changes due to the training on a human-manipulated dataset of graph variants. Additionally, to simplify the visualization of graphs, we conduct iterative division to obtain hierarchical clusters and visually map them to uncertainty-encoded nodes. At each level of granularity, topology-adaptive visualization is implemented to the coarsened graph and users can interactively explore graph connectivity in a top-down manner by expanding/folding the node of interest. To prove the effectiveness of our method, case studies were conducted on both synthetic and real-world datasets. The results show that our method not only helps users to understand the graph connectivity and query for similar patterns efficiently but also achieves high scalability for large graph visualization.