An empirical study on the impact of edge bundling on user comprehension of graphs

Abstract

Edges are one of the primary sources of clutter when viewing graphs as node-link diagrams. One technique to reduce this clutter is to bundle edges together based on a nearby source or destination. Combined with edge translucency, edge bundling is reported to reduce the clutter and reveal higher-level edge patterns. However there is very little empirical data on the impact of edge bundling on user performance, as well as the impact of graph characteristics such as edge density and graph size on the effectiveness of edge bundling as a graph-visualization technique. We have performed user experiments to evaluate the impact of bundling on user performance, using a set of randomly generated undirected compound graphs with varying vertex counts and edge densities. Our results indicate that edge bundling negatively impacts user performance at tracing paths between nodes, both in terms of accuracy and time. They also indicate that while edge bundling may provide no clear significant benefit in terms of accuracy for recognising higher-level cluster connectivity, it does provide a significant improvement in user response time.