Pairwise Visual Comparison of Directed Acyclic Graphs: A HCI Perspective

The goal of this project is to improve visualizations for the comparison of networks consisting of nodes and links. It is often challenging to detect differences between two or more such networks. The result of the project will be a set of guidelines which support such comparison processes in an appropriate way. This project will specifically concentrate on the visual comparison of directed acyclic graphs in node-link diagrams. Such comparison is needed, for example, in the analysis of phylogenetic trees in biology or in the assessment of contagion in financial networks. In these cases, the domain experts concentrate on finding commonalities and differences between the two networks being compared. This analysis is often undertaken in a visualized form. Visual exploration enables to identify where the differences are located and thereby to extract insights from these differences. Effective visual network comparison requires a well-designed visualization. Effective visualization uses guidelines, which are inter alia derived from research in cognitive psychology and human-computer interaction (HCI). Until now, HCI research has mainly focused on deriving guidelines for the visualization of single networks. There are still many open research questions concerning the comparison of two or more networks in node-link diagrams. The envisaged project results are novel guidelines for the visualization of network comparison, specifically for comparing directed acyclic graphs. The guidelines will improve visual comparison of networks, particularly applied in biology and financial stability analysis. We will identify a set of factors influencing detection and interpretation of graph differences in small and large graphs of various types (unlabeled, categorical and labeled). We will also present new benchmark datasets for visual network comparison studies in HCI. This project combines network visualization and cognition research. It therefore requires a close cooperation between experts from both fields. We thus propose a joint project in D-A-CH scheme, with two principal investigators (Dr. Landesberger von Antburg, Germany, network visualization) and (Prof. Margit Pohl, Austria, cognition and HCI).

The goal of this project was to analyze the design of diagrams showing networks (e.g. social networks, representations of relationships of elements in biology) so that the detection of similarities and differences between such networks can easily be supported. The networks are represented as so-called node-link diagrams consisting of nodes and links between them. Such comparisons are needed, for example, in the analysis of phylogenetic trees in biology or in the assessment of contagion in financial networks. In the research project, we conducted several evaluation studies to clarify these issues. The goal was to support developers to design usable representations that can be understood easily. These studies were influenced by insights from cognitive psychology, for example, by insights from research on human perception of similarities. Based on these studies, we could derive results that are relevant for the design of node-link diagrams. There is some research on similarity perception in information visualization, but not extensively so. Despite several existing visual network comparison approaches, the challenge of developing suitable techniques for visual network comparison nevertheless still remains open. One of the goals of our project was to clarify how to present the compared networks to the user so that differences and commonalities among networks can easily be spotted and their meaning can easily be recognized. This issue is complex as this also depends on the type of the networks (trees, directed acyclic graphs, undirected networks, labeled networks, etc.) and also on many other factors (design of nodes and links, density of nodes and links/white space, labels, ). One of our main goals was to identify the main influencing factors for the perception of similarity and differences. Based on our research, there is some indication that there are variabilities according to whether users search for similarities on the one hand or differences on the other hand. In the case of looking for similarities shape is the most important factor, in the case of looking for differences, factors like symmetry or edge crossings also play an important role. When looking for difference, users search the node-link diagrams sequentially for differences, while in the case of looking for similarities, a more holistic approach is chosen. Knowledge about the factors that influence the perception of similarities and differences is important for the design of node-link diagrams for comparison purposes.