Designing Visual Text Analysis Methods to Support Sensemaking and Modeling

Jason Chuang
PhD, 2013
This dissertation examines the co-design of interactive visualizations and statistical analysis algorithms to improve the process of visual analytics--to create effective workflows where human cognition and algorithms work in tandem to yield insights about large and complex data. I present the results of applying a human-centered iterative design process to a variety of projects: visualizations of statistical topic models, analysis tools to support topical quality assessment, a framework to support large-scale topical relevance assessment, and descriptive phrases for text summarization. My work has produced effective interactive visualizations, enabled more efficient analytic workflows, and contributed to our understanding of human categorization, topic modeling, and text summarization. I distilled design principles and design processes to inform practitioners on how to incorporate increasingly sophisticated models into data analysis tools. I designed, developed, and deployed various visual analysis tools for both builders and end users of statistical topic models; my approach led to not only improved visualizations but also the design of novel modeling techniques. I contributed survey methods and various datasets that can enable future studies on human-centered approaches to topic modeling. Across these projects, I demonstrate how we can effectively integrate diverse perspectives from information visualization, human-computer interaction, and machine learning to support effective model-driven data analysis.