Hybrid Visualization of Dynamic Multilayer Graphs – BLIZAAR

Our expert data forms what is usually called a complex system. A good model, as close as possible to the original system, is to consider it as a set of interconnected subsystems or layers. This model, which is at the core of our project, recently published and being popularized is called multilayer networks. To mine, navigate and visualize a multilayer network for getting more insight in the data, we use visual analytics. It consists in automatic analysis and interactive visualizations steered by domain experts. From a technological point of view, the experts want a platform for their personal use, to train new experts, for presenting or promoting their data to various types of audience. Thus, to these aims, we have modeled, developed and validated with domain experts a set of web-based tools despite the relative complexity of network-based interactive visualization on the web.

The main results of the project are:
- 4 working prototypes for visualizing and interacting with multilayer networks. These prototypes have been designed and evaluated in close collaboration with domain experts,
- for the application in life sciences, a methodology and tools for analyzing multilayer networks and enrich them with open-data sources,
- the organization of a high-level scientific seminar with the best international academics, researchers and domain experts of various application domains about the “Visualization of multilayer networks across disciplines”.

We contribute a lot to establish a new research field (star, Dagstuhl seminar, IEEE Vis workshop) in the international visualization community. We also propose tools which still need to be published and released properly. The project went well and the collaborations between the partners will not ceased with the end of the project. We hope to propose a new project on ML networks soon.
During the project, we all worked many times with domain experts from many application fields. We realize that multilayer networks are very easy to apprehend by domain experts and thus a good modeling tool even without visualization on a computer screen. The notion of layers seems quite natural for domain experts as long as the complex use of different types of edges (edges inside layers or between layers which may be of different kinds). This finding is not really an output of the project but a good motivation to continue working on ML network. There are many challenges left about ML network for instance about analytics (e.g., distribution patterns of edges between layers, centralities in the whole graph) and interactions.

The main scientific production is:
a detailed state of the art report about the visualization of multilayer network published in a famous international journal in the visualization field and publicly presented to the international community in visualization,
three papers published and presented in international conferences,
several communications presented orally or on posters in national or international events on specific technical aspects or in collaboration with domain experts.

BLIZAAR is an international collaborative project (PRCI) proposal which fits in the "Information and Communication Society" challenge. It involves French and Luxembourgish partners working in collaboration to craft novel ways of exploring and analyzing dynamic multilayer networks.

Our main application domain is digital cultural heritage with CVCE (Centre Virtuel de la Connaissance sur l’Europe). It concerns the collection, preservation, analysis and provision of open access to digitized cultural heritage objects, which may be any man-made object throughout history. In recent years, large amounts of digitized materials have for the first time become freely available to the public and scholars. The range of digital material covers many events and entity types, such political figures and institutions, which can be linked by many different relationship types that change over time. Domain experts often need to inspect various types of information from multiple perspectives to better analyze and grasp complex mechanisms. Our second application domain is life sciences in partnership with researchers from the environmental research and innovation department at LIST (Luxembourg Institute of Science and Technology). In recent years, the improvement in quality of practical experimental techniques has produced vast amounts of biological datasets representing many entities, such as genetic material, proteins and metabolites frequently modeled as multi-level networks. Biological systems are very complex and the levels rarely work in isolation. The experiences and problems in both application domains offer exciting new challenges and opportunities for the development of novel visual analytics (VA) solutions to provide useful tools to experts in each application domain.

Overall, existing VA approaches are limited and cannot be directly used to model complex real-world systems and fully capture their intricacy. Moreover, current visualizations of dynamic and multilayer networks do not convey temporal changes effectively. Thus, new interactive representations need to be investigated to better understand how such (potentially large-scale) networks are structured and evolve over time. The challenges are on the one hand that these networks may model complex phenomena involving multiple dimensions, which makes the analysis and tracking of changes difficult. On the other hand, although there are many ways to represent networks, experts still may not be sure which technique should be used in a given context.

Our first objective is to contribute novel graph visualizations that support the analysis and comprehension of dynamic multilayer network datasets. An important part of this objective is to build a clear understanding of the specific types of tasks that are applicable to multilayer network datasets. These tasks will need to be validated with end users and our novel visualizations evaluated in user studies. A second objective is to craft novel and interactive representations of dynamic and multilayer networks, by designing and combining visualizations. We will adopt an iterative process, mapping the problems of application domain expert users to multilayer dynamic network abstractions, then to interactive visualizations steered by domain experts. The proposal thus aims at expanding the state of the art on dynamic and multilayer network visualizations, by exploring, designing and evaluating new ways of representing these types of networks. We will contribute to (1) exploring and classifying visualizations of dynamic and multilayer networks including hybrid approaches, (2) designing novel interactive prototypes to explore some of these visualizations, and (3) evaluating these approaches in practice using case studies and user evaluations in the domains of digital cultural heritage and life sciences.