A User-Driven Interface for Exploring Visualizations Daryl H. Hepting and Paul Schmiedge University of Regina, Regina, CANADA ABSTRACT There is presently a variety of methods by which to create visualizations, and many of these require a great deal of manual intervention. Even with those methods by which it is easy to create a single visual representation, understanding the range of possible visual representations and exploring amongst them is difficult. We present a generalized interface, called c ogito, that permits the user to control exploration of the visualization output of various manual tools, all without the requirement to modify the original tool. Programming within the c ogito API is required to connect to each tool, but it is not onerous. We consider that the exploratory experience or activity is valuable, and that it is possible to easily create this experience for standard tools that do not normally permit exploration. We illustrate this approach with several examples from different kinds of manual interfaces and discuss the requirements of each. Keywords: exploration, visualization, batch, scripts, modular visualization environments, toolkits, human- computer interaction 1. INTRODUCTION When Haber and MacNabb 1 described the visualization pipeline, the question was how to realize a particular visual representation. Now, many years later, the question is how a user can find the particular visual repre- sentation that helps him or her gain some insight or communicate an idea. Sicard and Marck 2 found cognitive, didactic, and aesthetic logics in scientific pictures, which are not separable without knowledge of the author’s intent. For them, scientific pictures are “imbued with the ‘view’ of the author which claims to be objective. But, in fact, it is attached to ‘thought history’, technological history, scientific history and is marked by aesthetic choices, cultural bias, and perceptional practices.” All in all, the selection of a visual representation involves more than objective consideration of the problem to be depicted. Consider that any visual representation can be decomposed into parameters, each with their own values.A parameter could be “graph type,” with values including “bar chart,” “pie chart,” “line chart,” “scatter plot,” and so on. Each visual representation can be denoted as an N -tuple, where v i is a value of parameter P i . In practise, not all N -tuples may correspond to valid visual representations because of incompatibilities between values of different parameters. The Cartesian product of the values from all the parameters forms the N -dimensional space of available visual representations. 〈v 1 ,v 2 ,...,v N 〉∈ P 1 × P 2 × ... × P N The space of available visual representations can be very large, and it can be difficult to grasp the implications of all available combinations of values. This fact only exacerbates the problem of selecting and specifying the individual elements in a visual representation. In the midst of so many combinations, it can be difficult to find a visual representation that is apposite. The use of parameters and values to describe particular visual representations is an adaptation of Bertin’s 3 retinal variables, which he used to systematically explore marks on a plane and to show how those marks could be used to construct diagrams, networks, and maps. Similar concepts are found in Modular Visualization Environments 4 (MVEs) such as AVS 5 (Application Visualization System), and the toolkit philosophy of vtk . 6 Figure 1 presents a sample visual representation, based on a small dataset from Bertin 3 [page 100]. It provides a view of the French economy from the early 1960s. For each d´epartement in France, the data provides: the