November 22, 2011
12:00PM – 1132 Harold Frank Hall
Tobias Hollerer (chair)
Title: Interaction Methods for Large Scale Graph Visualization Systems – Using Manipulation to Aid Discovery
The amount of structured data available in our digital age is constantly growing and this calls for ways to analyze this data. One way for such analysis is through visualization, and structured data can typically be visualized as node-link graphs. Many tools are available for visualization of such node-link graphs. These tools often suffer from scalability issues, but more importantly they don’t offer meaningful interaction mechanisms for the user to explore the data. This dissertation work attempts to address both of these issues, by providing and evaluating scalable and useful interaction methods for graph visualization systems. In this dissertation we mostly focus on what we call manipulation methods, which are a set of interaction methods that enable the user to interactively manipulate the layout of a node-link graph. We provide a survey of existing interaction methods and introduce our own methods. These methods range from more specialized methods, such as for an interactive visual recommender system, to more general methods, which can be applied to any node-link data. We demonstrate the benefits of these interaction methods over pre-existing methods, and we also show the benefits of manipulation methods in general over visualizations that don’t allow layout manipulation. Additionally, we demonstrate how layout manipulation can be used as user input into an algorithm such as a recommender system.
In the presentation I will discuss our scalable web based graph visualization framework and present an evaluation of this framework, which enables the rest of this dissertation work. I will then present a motivating application for the need of scalable manipulation methods. I will also provide a survey of existing techniques before outlining our techniques. Then I will present the results of two user studies performed to evaluate the layout manipulation methods. The first study shows that for the tasks tested our proposed manipulation method outperformed the two other tested methods overall. The second study shows that our proposed manipulation method performs better overall than a system, which does not allow any manipulation of layouts. It also shows that with specific training, the users can perform even better at solving the tasks tested. I will then demonstrate how interactive layout manipulation can be used as user input into a visualized algorithm with an example from a visual recommender system. Finally, I will conclude the talk with a discussion of possible future directions.