Evaluating Visualizations of Sets and Networks that Use Euler Diagrams and Graphs
Almas Baimagambetov, Gem Stapleton, Andrew Blake and John Howse
Abstract.
This paper presents an empirical evaluation of state-of-the-art visualization techniques that combine Euler diagrams and graphs to visualize sets and networks. Focusing on SetNet, Bubble Sets and WebCola – techniques for which there is freely available software – our evaluation reveals that they can inaccurately and ineffectively visualize the data. Inaccuracies include placing vertices in incorrect zones, thus incorrectly conveying the sets in which the represented data items lie. Ineffective properties, which are known to hinder cognition, include drawing Euler diagrams with extra zones or graphs with large numbers of edge crossings. The results demonstrate the need for improved techniques that are more accurate and more effective for end users.
Nice research that shows that tools available for data visualization production need to be used with caution. They don’t always produce visualizations that are effective and efficient regarding their comprehension. The evaluation outcomes show that we need better data visualization tools.
I have two questions:
1. The evaluation outcomes of the research are based on the measurement of the visualization produced in terms of accuracy and effectiveness of properties. The authors point out that avoiding inaccurate and ineffective properties does not necessarily ensure effectiveness. In the same paragraph they state that there may be other factors, such as aesthetics, that can also impact effectiveness. How should I combine and reconcile these two statements?
2. What kind of empirical studies with human participants do the authors have in mind?
Hi Leonie,
Thank you for the feedback. Replies are as follows:
1. By “avoiding inaccurate and ineffective properties does not necessarily ensure effectiveness”, we mean that even if all of these properties are avoided, it is still possible that a diagram is not effective due to other factors, such as curve smoothness and curve-edge closeness. So, when a diagram has undesirable properties, then the diagram is likely to be ineffective, but the opposite (if no properties then effective) does not necessarily hold.
2. The studies are likely to focus on the categories of tasks most commonly performed on these types of diagrams, based on existing literature. The key variables to consider would be accuracy and speed with which the users will perform these tasks. The results of such studies will either reinforce the results in this paper, or will provide new insight with respect to the relationship between properties that were used in the evaluation and effectiveness of diagrams.