Visualizing Graphs as Maps with Contiguous Regions

Abstract

Relational datasets, which include clustering information, can be visualized with tools such as BubbleSets, Line-Sets, SOM, and GMap. The countries in SOM-based and GMap-based visualizations are fragmented, i.e., theyare represented by several disconnected regions. While BubbleSets and LineSets have contiguous regions, theseregions may overlap, even when the input clustering is non-overlapping. We describe two methods for creatingnon-fragmented and non-overlapping maps within the GMap framework. The first approach achieves contiguityby preserving the given embedding and creating a clustering based on geometric proximity. The second approachachieves contiguity by preserving the clustering information. The methods are quantitatively evaluated using embeddingand clustering metrics, and their usefulness is demonstrated with several real-world datasets and a fullyfunctionalonline system at gmap.cs.arizona.edu.