Reviewed by:
Michael
Stark
This paper describes the cone tree, a three dimensional structure used to represent hierarchical information. This work is done in the context of the Information Visualizer system, and is one of several visualization techniques that are tailored to specific kinds of data (for example, the Perspective Wall is used to represent linearly organized information). For a summary of a paper providing an overview of the Information Visualizer, click here.
The authors believe, and I agree, that their most important contribution (in the authors terms "clearest win") is the use of interactive animation to move some of the cognitive load to the human perceptual system. I think it is also important that they developed a technique for displaying more information on the screen, which should improve user awareness of navigation options.
The cone tree can be thought of as taking a two-dimensional drawing of a tree and folding the paper into a cone. A parent node is at the apex of the cone, and the child nodes are arrayed on the circle that forms the base. Each node on this circle is then in turn the apex of a cone at the next level. The basic means of navigating is to click on a desired node. The cone is then rotated so that this node is in the front, and the path from the root to that node is highlighted. The interactive animation capabilities from the Cognitive Coprocessor Architecture are used to animate the rotation of the cone tree, allowing the user to more easily track where he is in the hierarchy.
In addition to the animation, other techniques are used to aid navigability. A 3D perspective view of the tree gives the same effect as a fisheye view, without having to compute a degree of interest function. The connections between levels are translucent, and lighting and shadowing is used to improve 3D depth perception.
The authors motivate the use of 3D by showing a graph of aspect ratio against depth of tree for 2D trees wtih branching factors of two and three, and for a cone tree. The aspect ratio grows quickly in the 2D cases, and is fixed in the cone tree case. The tradeoff is that fixing the aspect ratio limits the depth of the cone tree (to 10 in their example). One question I have is whether aspect ratio is the best measure of how efficiently screen space is used.
The authors did some analysis of how the cone tree might be used in applications. This analysis did show that information could be stored in efficient ways, and that users could visualize things that they couldn't before. However, these examples left open the question of whether a user would really want to see these visualizations, or whether there are alternate approaches that are just as effective.
The paper finishes with a discussion of their results, reemphasizing there belief in the importance of interactive animation. They also discuss the limits on depth, branching factor, and number of nodes that can be effectively displayed, and compare with a 2D case. They do have the insight that rotating the cone can be considered a form of scrolling. However, in my mind the difference is that with the cone tree you see the structure of where you are scrolling to, although in less detail, where with 2D scrolling the information you are looking for isn't on the screen at all.
Questions