The application of computational modeling to data visualization

摘要: Researchers have argued that perceptual issues are important in determining what makes an effective visualization, but generally only provide descriptive guidelines for transforming theory into practical designs. In order to bridge the gap between and practice a more rigorous way, computational model of primary visual cortex is used explore perception data visualizations. A method presented automatically evaluating optimizing visualizations analytical task using human vision. The relies on neural network simulation early processing retina cortex. activity resulting from viewing information visualization simulated evaluated produce metrics effectiveness tasks. Visualization optimization achieved by applying these as utility function hill-climbing algorithm. This applied evaluation two types: 2D flow node-link graph visualizations. The various representations fields advection Laidlaw et al. predictive power examined comparing its performance subjects four types. results show same overall pattern humans model. both cases, best was obtained containing aligned edges. Flow done streaklet-based pixel-based parameterizations. An emergent property head-to-tail streaklet alignment, parameterization LIC-like result. The also diagram node connectivity two-layer diagrams. correlates with performance, terms accuracy response time. Node-link optimized diagrams exhibit aesthetic properties associated good design, such straight edges, minimal edge crossings, maximal crossing angles, yields empirically better task.