Collaborative Research: III: Medium: Towards Provably Effective Visualization Design

Data visualizations---from simple charts and graphs to complex models and dashboards---are used by millions to make sense of and communicate data. However, designing effective data visualizations requires a unique combination of skills, including statistics, graphic design, and expertise within a target domain such as medicine, public health, or engineering. This project aims to enable computers to serve as more reliable and robust assistants that provide guidance and feedback to help analysts make effective visualization design decisions. A central goal is to develop ``provably effective'' visualization analyses that can be tested against current “best practices”, theoretical models, and experimental data. This approach enables us to automate visualization design decisions that already have strong support within the scientific literature. By integrating these automated features into visualization tools, we can help thousands of analysts quickly navigate millions of data-driven decisions in their daily work. By uniting existing scientific theories under a single framework, we can also help researchers implement their findings within new and existing data visualization tools as well as rigorously test these tools to ensure they behave as intended. To reach this goal, the project targets three interleaved technical challenges, bridging reasoning about user goals and knowledge, formal visualization specifications, and actual visual output. The first track models a user's prior knowledge and goals as knowledge graphs to reason about visualization strategies. Given a task context, a formal space of visualization specifications can be searched to identify those that accord with both the task and perceptual design guidelines. The second track concerns specification-level visualization reasoning by incorporating richer notions of task and data developed in the first track, as well as by creating design knowledge bases via novel methods for identifying gaps and learning both design constraints and their weights. However, reasoning only about specifications stops short of the visual output that people see. In response, the third track develops an operational semantics of visualization to analyze and validate the effects of specification changes on graphical output. This approach enables deduction of new chart-level design constraints and systems-level optimizations. The project combines the results of these tracks into an integrated system for visualization design reasoning, which in turn will be applied to support scalable visualization and multi-view dashboard designs. Resulting tools and ideas will be disseminated through open-source software, tutorials, and visualization course curricula. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.