Data lineage and information density in database visualization

Visual representations of data help users interpret and analyze information. We have identified two key issues in existing visualization systems: data lineage and information density. This dissertation defines these problems and details solutions for them. We show that our techniques can be applied in database visualization systems, and we discuss how they improve the usability of these systems. The data lineage problem occurs when users apply a sequence of processing steps to input data sources; when viewing the final result, these users may wish to trace certain elements in the result back to the original input items. We call these types of queries data lineage queries. Current systems, e.g., geographic information systems or scientific visualization systems, provide little support for this task. In the first part of this dissertation, we discuss techniques for allowing users to access intermediate results efficiently while performing data lineage queries. We then introduce weak inversion and verification and show how they can be used to reconstruct the (approximate) lineage of derived data. Because they eliminate much of the irrelevant source data, weak inversion and verification can greatly reduce the amount of source data the end user must examine while performing a data lineage query. Visualizations often display too much information, making it difficult for users to interpret them. Similarly, visualizations often display too little information, thereby underutilizing display space. In the second part of this dissertation, we describe the general principle of constant information density. We show how both semantic and spatial transformations based on constant information density can be applied to create visualizations with appropriate density, thereby minimizing clutter and sparseness in the display. We describe an end-user programming environment in which users can construct visualizations with constant information density.