"Navigating Hierarchically Clustered Networks through Fisheye and Full-Zoom Methods" by Doug Shchaffer, Zhengping Zuo, Saul Greenberg, Lyn Bartram, John Dill, Shelli Dubs, and Mark Roseman

By Tammara Combs

The authors of this paper describe the difference between traditional full-zoom navigation techniques and the fisheye view. They discuss an experiment comparing full-zoom and fisheye methods for viewing hierarchically clustered networks. The results of the experiment showed that users completed tasks faster and required much less zooms during navigation using the fisheye method. At the end of the paper, the authors discuss some ideas on how they can make their variable-zoom method work for users of real systems. Some solutions and limitations are discussed and also a few guidelines for practitioners are suggested.

The fundamental problem that is being addressed in this paper lies in the question of, What is the best visualization strategy that will allow users to view hierarchically clustered networks while addressing the importance of balancing local detail with global context in a single view? The authors do an excellent job of addressing the problem right away. They also present a thorough explanation of previous research -- carefully pointing out research ideas borrowed and also limitations that they hoped to address.

More specifically, the authors wanted to answer a more immediate question about which is better: full-zoom methods or fisheye views. The problem with the full-zoom technique is that when users are zoomed out there is not enough detail, but when they are zoomed in there is not enough context. On the other hand, fisheye view techniques can be graphically distorted to show objects of interest to the user in full detail, while at the same time provide less detail for information further from the focus of attention. Solutions to the problems addressed by this research would be most useful to telephone network administrators, control room operators, and many others who use 2D networks and hierarchical data.

This paper was written, not to add a different view or perspective to fisheye views, but to do a quantitative analysis to assess just how useful the different views are when operators have to use them in real-time control environments. In the field of Human Computer Interaction, researchers so often make new tools, but there has been by far less time devoted to evaluating these tools to see if they are indeed helpful to the user. Therefore, I feel the authors should be commended for contributing the results of this thorough user study for other practitioners in the HCI community.

While this paper was well-written and explained, I have a few concerns about the validity of the results of the controlled experiment. This concern stems from the subjects used, the task ordering and the users’ preference of zoom. Perhaps the experiment could have been extended to more than users with computer science backgrounds. The order in which users performed the tasks could have produced a significant affect. Schaffer, et al. had half the users perform task A then task B with full-zoom. Then they had the remaining half perform task A then task B with the fisheye technique. Perhaps performing task B first would have produced different results. The last concern, the users’ preference of view of zoom could be handled in two ways. The user could simply be asked which zoom-method he/she preferred or they could choose which zoom-method they want to use given certain tasks.

There was an issue that appeared throughout the entire paper, and that is the trade-off between distortion and available screen space. However, the authors never state this. Instead it is only alluded to. I think this would be a good point for the authors to scrutinize.

There is also one other factor I feel the authors failed to discuss the fundamental issue of scalability. The authors tell us how many nodes were used in the experiment, but just how many nodes can appear in either view before the system are rendered unusable?

I believe the authors are addressing these issues by continuing the testing of their system by way of user studies. Perhaps the feedback that users give will aid in building a much more useful system. It is understandable that the authors could not address every issue, but they have done a wonderful job presenting their work. The comments written here are merely suggestions for future studies or research.

How were the instructions administered to the subjects?

Can the fisheye view system be used with any hierarchical data?

How would you represent a node or sub-cluster that belonged to more than one cluster?