Towards Better Perception of Graph Visualization

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering


PhD Thesis Defence


Title: "Towards Better Perception of Graph Visualization"

By

Mr. Yong WANG


Abstract

Graph data is ubiquitous in lots of application areas such as social media, 
biological networks, financial transactions and software engineering. To 
help users understand and analyze those graph data, the visualization community 
has been actively working on graph visualizations. Various graph 
visualization methods have been proposed in the past decades. However, due to 
the limited screen space, the unavoidable trade-off of different aesthetic 
criteria, human visual perceptual capability limit and others, users are not 
able to easily gain a comprehensive and accurate perception of graph 
visualization in all the situations, especially when the graph size increases. 
In this thesis, we propose novel approaches to enhance the user perception of 
both static and dynamic graph visualization.

For static graph visualization, prior studies have proved that it is impossible 
to optimize all the aesthetic criteria simultaneously. Ambiguity and other 
misleading information may always exist in the graph layout results. To provide 
users with an accurate and comprehensive perception of graph visualizations, 
we propose AmbiguityVis, a novel approach to inform users of the 
potential perception problems in the graph layout. More specifically, new 
readability metrics are proposed to quantify the ambiguities and heatmap-based 
visualizations are present to visualize those ambiguities. For dynamic 
graph visualization, we aim to enhance the perception of two major 
visualization ways of dynamic graphs, i.e., animation and small multiples. We 
first propose a vector field design approach to improve animated transitions of 
clustered objects. It explicitly enhances coordinated motion and avoids 
crowding, better supporting the tracking of individual objects and communities 
in a scene. Then, considering that the common uniform timeslicing can generate 
cluttered timeslices when edge bursts occur and empty timeslices when few 
interactions are present, we introduce a nonuniform timeslicing approach based 
on histogram equalization for small multiples. It divides the whole time range 
in a non-linear way and strikes a balance between temporal distortion of 
time dimension and similar visual complexity across intervals.


Date:			Monday, 30 July 2018

Time:			2:00pm - 4:00pm

Venue:			Room 3494
 			Lifts 25/26

Chairman:		Prof. Xueqing Zhang (CIVL)

Committee Members:	Prof. Huamin Qu (Supervisor)
 			Prof. Pedro Sander
 			Prof. Chiew-Lan Tai
 			Prof. Fugee Tsung (IEDA)
 			Prof. Min Chen (Oxford Univ.)


**** ALL are Welcome ****