An intelligent system for task-specific visualization assistance

In analyzing visualized data, the scientist interprets and draws conclusions from the information that is displayed. Arriving at a solution can tax the scientist's limited attention and short-term memory. Visualizations have the potential to improve problem-solving performance by shifting the burden from those cognitive processes that are severely limited, such as working memory, to cognitive processes that are less limited, such as perception and pattern recognition.; One approach to reducing this cognitive load is to provide intelligent assistance in the visualization system. Visualization Assistant (VA), described in this dissertation, supplies such intelligent assistance by constructing effective representations that avoid overloading the scientist's perceptual and cognitive systems. One method for reducing cognitive load is to shift some information processing from attentive to preattentive processes. Since early processing occurs without direct, conscious attention, the cognitive requirements of a task may be substantially reduced. VA shifts information processing from attentive to preattentive processes through the use of design constraints. These constraints, based on results from empirical studies of visual and graphical perception, limit the number of potential designs to those that improve the scientist's task performance by minimizing unnecessary perceptual processing. The detection of simple properties, perceptual grouping, and texture all occur early in perceptual processing. The circumstances under which these operations occur preattentively provides knowledge about ways to transform some of the task's information processing from slow-serial to fast-parallel operations.; In designing effective data visualizations, an understanding of the scientist's task is crucial. Task-dependent designs enable the scientist to efficiently and accurately decode the necessary information from the visual display. Since scientific data sets are quite complex, VA develops a taxonomy of tasks that can handle large sample sizes and multiple dimensions.; VA uses case-based reasoning for storing designs as cases, retrieving the ones consistent with the constraints, and ranking the alternative designs according to their relative effectiveness. With case-based reasoning, designs are generated once, stored as cases, and assigned indices for subsequent retrieval. Reusable designs reduce the time in which effective designs become available for satisfying the scientist's task. As new combinations of tasks are encountered, VA acquires the additional knowledge by adding the composite designs to the case base. The composition process uses knowledge, stored in the form of composite constraints, to avoid interactions which may adversely affect task performance by interfering with the ability to focus attention on relevant objects. The system then guides the scientists by ranking designs in the browser according to their relative effectiveness. Although these ranked designs suggest possible solutions, the selected solution must first be adapted to fit the new problem exactly. Following adaptation, the system's attention director directs the attention of the scientist and provides guidance in using the graphical representation. By communicating directions, the attention director guides the scientist in more deliberate searching of the rendered visualization.; VA responds to the scientist's evolving goals by enabling the scientist to specify tasks interactively by using the visualization's current context as a starting point for adapting designs. Although VA supplies perceptually effective designs consistent with the scientist's task specifications, the locus of control remains with the scientist, who maintains responsibility for all decision making.