Efficient creation and analysis of visualizations for data communication

Abstract

The influential rise of the Internet and social media has accompanied ever-increasing public needs for accessing information and data. In light of this trend, data visualizations have emerged as one of the primary mediums for public data communication. Thus, a large number of visualizations have been produced and shared on the web, raising new challenges and problems in both society and academia. In this thesis, I investigate this phenomenon and accompanying problems through a careful combination of research methods including literature survey, empirical studies, and machine learning. Specifically, this thesis focuses on:   (1) Building novel recommenders for authoring high-quality visualizations. Many web visualizations are made by non-experts and suffer from quality problems such as poor readability and insight. I present three automated approaches to assist the general public in creating visualizations, including MobileVisFixer for generating mobile-friendly designs, LQ2 for authoring aesthetic layouts, and MultiVision for designing analytical dashboards.   (2) Formalizing visualizations as a new first-class object for efficient analysis. Through a comprehensive literature survey into ten research fields in computer science, I argue that visualizations are becoming a new data object like images and text. I further formulate the emerging research field as visualization processing and analysis that concerns processing digitized visualizations and extracting meaningful information. My work Computable-Viz presents a formalism for operating on multiple visualizations, thereby creating novel applications such as interactively combining visualizations in AR environments. By integrating those two perspectives, this dissertation contributes to a new online knowledge ecosystem – both analyzing web visualizations to distill knowledge and assisting the public in producing new visualizations to communicate data. I hope that this ecosystem will continue stimulating new theories, problems, techniques, and applications to further bridge the public with data.

Date of Award	2022
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology
Supervisor	Huamin QU (Supervisor)