HCC: III: Database Visualization Theory and Practice

Data visualization tools like Tableau and Excel have transformed how scientists, businesses, and the public make sense of information. But these tools share a hidden limitation: they can only work with data that has been manually flattened into a single spreadsheet-like table. In reality, most data is stored in relational databases — collections of linked tables that capture rich relationships, such as how patients connect to diagnoses, or how species connect to environments. When analysts flatten this data to visualize it, they discard the very structure that gives it meaning, risking misleading charts, hidden errors, and conclusions that don't reflect reality. This project builds a new scientific foundation for data visualization that works directly with relational databases, ensuring that what you see faithfully reflects what is in the data. The result will be a new class of visualization tools that are more correct, more expressive, and more accessible — lowering the barrier for scientists, policy makers, and the public to explore complex data without specialized technical expertise. This project develops the theory and systems for database visualization — a new paradigm that extends declarative visual grammars to treat relational databases, including their schemas, keys, and foreign key constraints, as first-class inputs. The core intellectual contribution is a formal theory that defines visualizations as constraint-preserving mappings from relational structure to visual structure, ensuring soundness: that the visual representation faithfully encodes the data's relationships and integrity. The research produces three interconnected artifacts: (1) a compositional algebra over multi-view visualizations that supports constraint-aware transformations and symbolic reasoning; (2) a JavaScript visualization library (dvl) that operationalizes the theory for in-browser databases, extended via a compiler that translates visual specifications into optimized, incremental SQL pipelines for large-scale data; and (3) a Visual Data Pool Analytics system that allows analysts to interactively curate and transform a pool of relational tables while the system automatically maintains a sound, up-to-date visualization. The work is evaluated through system benchmarks, controlled user studies with visualization designers and data analysts, and real-world deployments with collaborators in biology, disaster preparedness, and data science tooling. 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.