Designing Automated Assistants for Visual Data Exploration

Visual data exploration enables analysts to identify trends and patterns, generate and verify hypotheses, and detect outliers and anomalies. However, the overwhelming number of decisions required in visual data exploration presents a barrier to discovering useful, action-able insights from data. To address this challenge, in this dissertation, we investigate how automated assistance via tooling aids visual data exploration.

We introduce four systems to survey the design space of visual exploration assistants across different analytical tasks and interface modalities. We first describe VisPilot and Zenvisage++, two novel visual exploration assistants that accelerate the data exploration process for individual visual analysis tasks: drill-down analysis and pattern search. Next, we examine visual exploration assistants aimed at supporting multiple types of visual analysis tasks. We introduce Frontier, a general-purpose visual exploration assistant within a GUI-based charting tool that recommends potential next steps in a mixed-initiative visual analysis workflow. We further develop Lux, a general-purpose visual exploration assistant situated within a computational notebook that provides proactive, always-on recommendations within an exploratory programming workflow. Findings from this dissertation contribute towards designing an intelligent visual exploration assistant that suggests helpful tailored feedback based on user’s analytical needs and seamlessly guides users towards data-driven insights.