[Back]

B. Alsallakh:
"State-of-the-art methods for visualizing set-typed data";
Talk: Seminar of the Computational Intelligence Group, University of Kent, Canterbury (invited); 2013-12-09.

A variety of information about real-world entities can be modeled as a system of sets that are defined over these entities. As example, such a system can model which hobbies (sets) are practiced by a group of people (elements), or which classifiers (sets) correctly classified which data samples (elements).
Analyzing how multiple sets overlap and finding out which elements belong to which sets and overlaps are essential tasks that arise when dealing with set-typed data, i.e. data that involve element-set memberships beside other element attributes. As example, a machine-learning expert might want to find out which samples were correctly classified by all classifiers, by a certain classifier only, or by two or three certain classifiers only. While Venn and Euler Diagrams are by far the most common and natural visual representations for set overlaps, they can handle only a small number of sets, which limits their applicability in real-world problems involving a larger number of sets.
In this talk, I will explore a number of alternative Information Visualization techniques that have been developed over the past decade to visualize overlapping sets. After explaining the design rationales behind these techniques, I will discuss which data and tasks they are suited for, and what practical limitations they have. I will elaborate on a recent technique called Radial Sets by presenting two usage scenarios in the classification domain: analyzing multi-label classifications, and comparing the results of multiple classifiers.

multi-label classification, set-typed data, visualization

Project Head Silvia Miksch:
CVAST: Centre for Visual Analytics Science and Technology (Laura Bassi Centre of Expertise)