PhD Dissertation Defense by ZHANG Songheng | Usage-Aware Automated Visualization Systems

ABOUT THE TALK

Automated visualization systems reduce the expertise required to create charts by generating visual representations from input data. Despite advances in recommendation accuracy and generation quality, these systems share an unexamined assumption: that the pipeline from input to output is usage agnostic, and that deployment context imposes no design constraints on output. This dissertation argues that assumption is wrong, and asks: how can we characterize the design constraints imposed by distinct usage scenarios and incorporate them into automated visualization generation?

We demonstrate this across three scenarios, each representing a different input modality, user population, and physical context. In layperson analytical usage, users without visualization expertise rely on recommendation systems to analyze tabular data. AdaVis reframes recommendation as a one-to-many mapping using a knowledge graph with box embeddings, and generates natural language explanations for each candidate chart. Evaluations show it recommends multiple valid visualizations with explanations judged correct and useful by both novice and expert users.

In public mobile display usage, people view sensitive personal data on smartphones where bystanders can observe the screen. A perception-driven system adjusts spatial frequency and luminance contrast so charts remain legible at close range while becoming uninterpretable at distance. Two user studies (n=16, n=18) demonstrate comparable legibility at 30 cm and substantially reduced interpretability at 90 cm.

In text-based usage, visualizations are generated from natural language documents containing linguistic uncertainty markers. UncertaintyVis classifies uncertainty into four categories from corpus analysis and maps each to chart-specific visual encodings. A user study (n=12) shows 85% chart-to-text matching accuracy, reduced perceived cognitive demand, and 75% participant preference for uncertainty-preserving charts.

Together, these systems establish that usage scenario awareness produces outputs generic pipelines cannot achieve, and that it is a necessary and productive design principle for next-generation automated visualization research.

ABOUT THE SPEAKER

Songheng ZHANG is a PhD candidate in Computer Science at the School of Computing and Information Systems at SMU, under the supervision of Associate Professor Tony TANG. His research primarily focuses on data visualization recommendation and mobile data visualization.