Design and Evaluation of Visual Summaries to Improve Readability of Large Network Visualizations
Date
2025-07-04
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
ORCID
Type
Thesis
Degree Level
Masters
Abstract
Node-link visualizations are commonly used to gain insights into large network data where the entities of the networks (nodes) are represented as points, and relationships (edges) are drawn as straight-line segments or links. With the growing access to network data and visualization tools, such visualizations are increasingly appearing in infographics and documents intended for non-specialist readers. This necessitates understanding how these visualizations are perceived by end users who are not necessarily domain experts, and determining how visual summaries can be provided to ensure consistent interpretation of the displayed information. In this paper, we investigate the interpretability of node-link visualizations of large graphs. Despite the popularity of node-link layouts, little work has systematically studied how intuitive or reliable these representations are for everyday users when graphs become large and complex. Understanding such layouts' user interpretation is therefore critical for designing effective visualizations. We designed intuitive summaries that could be provided alongside the visualization to improve interpretation and evaluated these designs through two user studies with 20 participants each. We used real-life datasets such as online social networks and author collaboration networks for the studies. We automatically detected clusters using the HDBSCAN tool and filtered the top 50 high-quality clusters by alignment with modularity-based clustering. We also investigated the extent to which participants agreed on the identification of clusters in ForceAtlas2 visualizations and whether the automatically detected clusters align with those identified by participants. Our results indicate that the information perceived from traditional node-link representations can vary substantially, especially when the nodes are uniformly distributed rather than forming obvious visual clusters or tangled structures. When the clusters appeared as clear tangled structures, agreement was high and when the clusters were less visible in one dataset, agreement dropped substantially. We also observed that visual summaries greatly enhance the readability of these visualizations -- summaries that effectively reduce clutter were preferred by participants and were more accurate for typical interpretation tasks.
Description
Keywords
Human-centered computing, Graph drawings, Visualization design,
Citation
Degree
Master of Science (M.Sc.)
Department
Computer Science
Program
Computer Science