Finding a spatial mapping for an abstract data set that is cognitively useful for a specific task is a considerable design challenge. Information visualization draws on ideas from several intellectual traditions, including computer graphics, human-computer interaction, cognitive psychology, semiotics, graphic design, cartography,and art. The synthesis of relevant ideas from these fields with new methodologies and techniques made possible by interactive computation has led to the emergence of the infovis field over the past several years.
Questions about visual encoding are even more central to information visualization than to the somewhat older field of scientific visualization.The subfield names grew out of an accident of history and have some slightly unfortunate connotations when juxtaposed: information visualization isn't unscientific, and scientific visualization isn't uninformative.
Although not all of us agree on the distinction between the two, the definitions I'll use are that information visualization hinges on finding a spatial mapping of data that is not inherently spatial, whereas scientific visualization uses a spatial layout that's implicit in the data.
Many scientific data sets have naturally spatialized data as a substrate. For instance, airflow over an airplane wing is given as values of a 3D vector field sampled at regular intervals that provides an implicit 3D spatial structure. Scientific visualization would use the same 3D spatialization in a visual representation of the data set, perhaps by drawing small arrows at the spots where samples were taken, pointing in the direction of the fluid flow at that spot, with color coded according to velocity. Scientific visualization is often used as an augmenation of he human sensory system, by showing things that are on timescales too fast or slow for the eye to perceive, or structures much smaller or larger than human scale, or phenomena such as x-rays or infrared radiation that we can't directly sense.
In contrast, a typical information visualization data set would be a database of film information, with the title, length, year of production, and genre for each film. Such a data set is more abstract than the fluid flow example because there's no underlying spatial variable. One possible spatialization would be to show a 2D scatterplot with the year of production on one axis and the film length on the other, with the scatterplot dots colored according to genre [1]. This choice of spatialization is an explicit choice of visual metaphor by the visualization designer and is appropriate for some tasks but not for others.
André Skupin presents a visualization system for document repositories using self-organizing maps. Although his data set of conference paper abstracts is nongeographic, cartographic principles inform his choice of spatial representations. Skupin uses hierarchical clustering to create scale-dependent spatializations that have information densities appropriate for a given level of abstraction.
Robert Kosara, Silvia Miksch, and Helwig Hauser present further results about semantic depth of field, a visual emphasis technique that was originally presented at InfoVis2001. They present several sample applications where they direct the user's attention to the relevant objects by blurring the currently irrelevant objects in the scene.
Robert Erbacher, Kenneth Walker,and Deborah Frincke tackle the application domain of intrusion detection, providing a concise visual interface that lets system administrators inspect log files for evidence of machine misuse. Ugur Dogrusoz and his colleagues from Tom Sawyer Software offer a look at the many applications of graph layout and interactive editing software. Tom Sawyer is one of the success stories of infovis: the company has been at the forefront of technology transfer from research to industry since its inception 10 years ago.
Readers may contact Tamara Munzner by email at tamara.munzner@compaq.com or on the Web at http://graphics.stanford.edu/~munzner.