| In recent years, large multidimensional databases, or data warehouses, have become common in a variety of commercial and scientific applications. It is not unusual for these data warehouses to contain billions of tuples, each categorized by tens or hundreds of dimensions. A major challenge with these databases is to extract meaning from the important data they contain: to discover structure, find patterns, and derive causal relationships. A promising technique for the analysis of these multidimensional databases is visualization. To make visualization effective in this context, we need to develop tools that tightly integrate visual presentation and database queries, support interactive refinement of the display, and can visually present a large number of tuples and dimensions.; This dissertation introduces a formal approach to building visualization systems that addresses these demands. The foundation of the dissertation is the Polaris formalism, a language for precisely describing a wide range of table-based graphical presentations of relational information. A key aspect of this formal language is the ability to compile visual specifications automatically into the precise queries and drawing commands necessary to generate the display. This ability enables us to design systems that closely integrate analysis and visualization. Using the Polaris formalism, we have built two interactive systems: the Polaris interface and a framework for multiscale visualization.; The Polaris interface for the exploration of multidimensional databases extends the popular Pivot Table interface to generate a rich, expressive set of graphic displays. The Polaris interface is simple and expressive because it is built upon the Polaxis formalism. Analysts can incrementally construct complex queries, receiving visual feedback as they assemble and alter the query. The Polaris interface is a generally applicable tool that tightly integrates analysis with visualization. This dissertation also demonstrates how to use the Polaris formalism and data cubes to specify and implement domain specific multiscale (pan-and-zoom) visualizations efficiently. The presented approach to multiscale visualization addresses several limitations in the current approaches by introducing multiple zoom paths into the data and providing general mechanisms... |
