Visualization research spans a remarkable range of scientific disciplines and corresponding visualization techniques. Visualization researchers have discovered that certain operations are needed across this entire range. These operations include comparing visualizations of two different datasets, as well as performing algebraic operations on two or more visualizations, such as visualizing the difference between two datasets. Furthermore, the need to explore multiple visual representations simultaneously arises especially in information visualization, because different techniques often extract different visual features and the complexity of the data. A visualization spreadsheet is an excellent way to address these issues that involve multiple visualizations.
Over the past year we have learned that the spreadsheet approach is a powerful and intuitive technique for interacting with 3D information visualizations. In this paper, we showed that a visualization spreadsheet supports information visualizers as they are confronted by the challenge of visualizing a wide variety of different types of data. Two types of interaction tasks are important to the users. One is being able to quickly prototype an application for interacting with data. The other is being able to apply operations such as compare/contrast, rotation/translation/zooming, filtering/sorting/searching and other domain-specific operations. The visualization spreadsheets described in this paper have proven useful in the above two tasks in the domains we examined--visualizing sequence similarity data in molecular biology, two sets of time-series matrices, and visualizing the steps in the 3D Delaunay triangulation algorithm.
For each domain, we showed how our prototype spreadsheets enabled users to compare visualizations in cells using the tabular layout. Using these domain examples, we also showed how users use the spreadsheet to display, manipulate, and explore multiple visual representation techniques for their data. By applying different operations to the cells, we showed how visualization spreadsheets afford the construction of 'what-if' scenarios. The possible set of operations that users can apply is now a rich set of domain-dependent as well as domain-independent operators, such as animation, filtering, and algebraic operators between cells. Subtraction between cells, for example, can be applied both at the pixel level as well as at the geometric-object level. Other operations may now be coordinated, such as applying the same rotation manipulation across a group of cells.
We also examined the differences between two interaction styles--command language and direct manipulation. Our first prototype for visualizing sequence similarity data uses the noun-verb direct manipulation model. We find that this interaction style is somewhat less flexible than a command language, but supports most of the needed flexibility in an easy-to-use framework for a specialized domain. The command language used in the second prototype is based on Tcl, and is considerably more flexible for users to define their own macros and modules. It also allowed expert users to quickly perform a number of complex operations.
The spreadsheet approach is a powerful and intuitive technique for interacting with the information visualizations in a structured way. Further research is needed to understand the properties of visualization applications that work well in spreadsheets, to investigate the appropriate user interfaces at all levels, and to develop a framework to enable rapid development of visualization spreadsheet applications.