This project studies methods for assisting with the navigation of large, complex information spaces. Our goal is in the construction of a navigation system designed to help viewers visualize, explore, and analyze large, multidimensional datasets.
Our technique combines a detailed local display and a high-level global overview of the locations and structure of areas of interest within the dataset. Our local view uses perceptual cues to harness the abilities of the low-level human visual system. The global overview is built in two separate stages. First, elements of interest are identified using a combination of: (1) explicit rules provided by the viewer, and (2) implicit rules built by watching what viewers select, where they move, and what they examine. Next, the elements are clustered into one or more areas of interest. We are investigating the use of graph construction techniques like planar triangulations and minimum spanning trees to link the elements together. We seek an underlying graph that: (1) supports efficient navigation via the application of graph traversal algorithms, and (2) provides an effective global overview to visualize the areas of interest and the relationships that exist between them.
Datasets from the oceanography and e-commerce domains are being used to test our system in a practical, real-world environment. Experiments with domain experts are being used in part to provide anecdotal feedback on our system, and in part to identify fundamental navigation and exploration tasks performed during visualization. These tasks will then be integrated into a controlled experiment that studies the performance of our system vis-à-vis a system without navigation aids, and existing focus+context visualization techniques specifically designed to display these types of large, complex datasets.
Brent Dennis (PhD candidate)