Dagstuhl Seminar 19061
Visual Analytics of Multilayer Networks Across Disciplines
( Feb 03 – Feb 08, 2019 )
- Nathalie Henry Riche (Microsoft Research - Redmond, US)
- Mikko Kivelä (Aalto University, FI)
- Fintan McGee (Luxembourg Inst. of Science & Technology, LU)
- Guy Melançon (University of Bordeaux, FR)
- Tatiana von Landesberger (TU Darmstadt, DE)
- Andreas Dolzmann (for scientific matters)
- Simone Schilke (for administrative matters)
- Human Factors and Multilayer Networks : article in Workshop on Visualization of Multilayer Networks (MNLVIS '19) at IEEE VIS '19, October 21, 2019, Vancouver, BC, Canada, 2019 - Pohl, Margit; Kerren, Andreas - Los Alamitos : IEEE, 2019. - 4 pp..
- Layer Definition and Discovery in Multilayer Network Datasets : article - MacGee, Fintan; Morin, Ludovic; Stefas, Mickael; Zorzan, Simone; Ghoniem, Mohammad - MLNVIS2019, 2019. - 5 pp..
- Layer entanglement in multiplex, temporal multiplex, and coupled multilayer networks : article - Skrlj, Blaz; Renoust, Benjamin - Cornell University : arXiv.org, 2020. - 32 pp..
- Patterns of Multiplex Layer Entanglement across Real and Synthetic Networks : article - Skrlj, Blaz; Renoust, Benjamin - Cornell University : arXiv.org, 2019. - 12 pp..
- Visual Analysis of Multilayer Networks - McGee, Fintan; Renoust, Benjamin; Archambault, Daniel; Ghoniem, Mohammad; Kerren, Andreas; Landesberger, Tatiana von; Melancon, Guy; Otjacques, Benoit; Pohl, Margit; Pinaud, Bruno - San Rafael : Morgan & Claypool Publishers, 2021. - 150 S. - (Synthesis Lectures on Visualization ; 8, 2021, 1). ISBN: 978-1-636-39143-4.
Networks, used to understand systems, are frequently modelled, analysed and visualised as a single complete network. However, real-world systems are often more accurately modelled as a set of interacting networks, or layers. These so-called Multilayer Networks are studied by researchers both in network visualization and in complex systems — the domain from which the concept of Multilayer Networks has recently emerged. Moreover, researchers in various application domains study these systems, e.g. biology, digital humanities, sociology, business and financial analytics, geography, and journalism. This Dagstuhl Seminar will bring participants from these various areas together.
Visual analytics focuses on the design of visual representations of data that properly supports analytical tasks so that it reveals insight present in the data used to conduct the analysis. The challenge is to draw the right image based on sound statistics computed on the data that provide useful answers to domain centred questions. Multilayer Networks provide new challenges, in terms of visualization, analytics, interaction, and modelling. New tasks and domain problems are exposed across the application domains, which are not addressed by existing visual analytics approaches.
Layers can model many different concepts across application domains, and often terms such as heterogeneous, multi-faceted, multi-modal, or multi-relational networks, amongst others, are used to describe networks which can be modelled as multilayer networks. The core topics of the seminar will include the concept of layer, the novel visualization research avenues that will need to be exploited, and the required analytics techniques to extract salient properties and patterns in multilayer data. We will also examine the requirements and challenges in the application domains, as these drive each of the other topics.
The purpose of this seminar is to create an interdisciplinary community of researchers and practitioners of different disciplines to discuss challenges and outline research directions for visual analytics of multilayer networks. The seminar aims to provide:
- real-world use cases (datasets and tasks) and open problems in a wide range of disciplines to inform and inspire visual analytics researchers and practitioners.
- cutting-edge interactive visual exploration and analytics tools to inform and outline promising solutions to application domain researchers.
- discussion leading to novel visual analytics techniques and novel applications of existing ones to push the state-of-the-art forward both in visual analytics and application domains.
The topic of multilayer networks has recently emerged from the field of complex systems, however many of the of the fundamental concepts and ideas have existed for some time, in fields such as sociology, and often under different nomenclature, such as multimodal, heterogeneous or multiplex networks. The multilayer network framework of Kivelä et al  has collected many of these concepts and different labels, along with example data sets, allowing us to recognize the multi-disciplinary importance of multilayer networks as a topic. Despite the importance of this topic, it is only recently that the visualization community is beginning to consider approaches for the visual analytics of multilayer networks. This seminar was the first to bring together practitioners from multiple domains to discuss the visual analytics of multilayer networks. These fields included data visualization, complex systems, digital humanities, biological sciences, health informatics, and sociology. The primary goal of this seminar was to bring together these thinkers and practitioners from different disciplines to drive forward new advances on the topic. The seminar was designed to foster discussions between researchers and designers of visual analytics tools, those who define the underlying theory, and the the end-users of these tools. To push research further and produce significant impact in industry and general public practices, the research community needs to establish a deeper collaboration between data scientists and researchers from applications domains (e.g. biologists, social scientists, business analysts, journalists, physicists), who collect and analyze the data; and researchers in maths, physics and computer science who push the state-of-the-art, producing visualization and analysis models, algorithms and tools. This deeper collaboration starts with building an understanding of the needs and tasks of network analysts. This seminar was an important first step, leveraging cross domain synergies with the goal of identifying the shared underlying problems and helping to solve them. The domain experts presented their domain problems early on in the seminar, and then interacted with two different sets of visualization experts in two separate breakout sessions. The motivation for this was to expose the visualization experts to many different domain problems and to expose the domain experts to multiple approaches to their problems. Our goal was to not only to advance research in the field of visualization, but also to provide techniques to help the domain experts to advance research in their own field. Interdisciplinary intersection was a key part of the methodology of our seminar.
The seminar featured talks and working groups that discussed topics on visualization, analysis, theory and applications of multilayer networks (see Sections 3 and 4). The application domain focus was maintained throughout the seminar. Experts from application domains gave talks in the first day and a half highlighting the problems they encountered. Then there were two breakout sessions where each experts was assigned a different group of visualization experts, allowing the domain experts to brainstorm solutions to their problems with different sets of visualization experts.
The talks brought the interdisciplinary participants initial information on a) current application problems dealt with in the area of multilayer networks and b) current visualization, analysis and systems solutions.
The purpose of talks by application experts was to make sure that the potential solutions provided by the interactive visualizations and analytics fully meet the requirements of those who actually use them, i.e. the system biologists, social network analysts, historians, etc. Therefore, the talks provided understanding of the data and problems/tasks/goals when analyzing multilayer graphs by the domain experts. The talks covered application areas of social networks by A. Cottica and by M. Magnani, information circulation in an international organization by M. Grandjean, digital humanities by M. Düring , multi-omics data by S. Legay, population health by M. McCann, digital ethnography by A. Munk(see sections 3). These talks allowed the visualizations and complex systems theory experts to gain some insight into the domain experts problems. As we also wanted the domain experts to be exposed to multiple approaches to their problems, we had two breakout sessions after all of the domain experts presented their personal topics. In these breakout sessions, each domain experts was assigned a small group of visualization researchers to further brainstorm, mapping visualization problems to domain problems. Different researchers were assigned to each domain expert for each session. This exposed the domain experts to multiple visualization approaches, and allowed for synergies between application domain problems to be identified by the visualization researchers. At the end of each breakout session, each groups gave a short report back to other participants, allowing for further discussion and cross fertilization of ideas. This approach ensured that that both the domain experts and the visualization experts had a wide range of ideas to explore as part of the working groups in the later half of the seminar.
The purpose of talks by visualization, analysis and systems experts was to present currently available solutions to multilayer network visualization and analysis (see also Section 3). These talks were dispersed throughout out the week. The talk topics were: an introduction to multilayer networks by F. McGee, a complex systems perspective on the concept of multilayer network by M. Kivelä, survey of multilayer visualizations by G. Melancon, Py3plex library for visualization by B. Skrlj, interaction with multilayer network visualization by B. Renoust. This allowed application experts to get to know the advantages and limitations of existing solutions. The talk schedule was flexible, for example, due to a high level of interest form all attendees M. Kivelä gave a second question and answers session to his talk the following day.
At the midpoint of the week we defined the working groups. The breakout session stimulated a large amount of discussion and ideas across participants of all disciplines. Following on the breakout sessions discussions, all seminar participants wrote down topics and ideas that were that were of interest to them of pieces of paper, which were affixed to a board. Similar topics were re-positioned closer together on the board, until all participants reached a consensus of five topic areas for discussion within working groups. The resulting working groups were as follows:
- Unifying Terminology and visual analytic approaches: One open problem of multilayer network analysis and visualization is the inconsistent terminology across disciplines. There are many different names given to networks with such characteristics, outlining the current lack of consistent definitions between disciplines, such as heterogeneous, multi-faceted, multi-modal, or multi-relational networks, amongst others (see ) and in the vast majority of cases it is possible to model them as multilayer networks. The discussion group assessed various types of networks from visualization, application and systems perspective. It discussed possible unification of these perspectives in one visual analytics framework and identified open challenges (see Section 4.3).
- Analytics, Communities Comparison and attributes: Visual analysis of multilayer networks is also concerned with the exhibition of salient properties and patterns in data. Salience in networks is often captured through metrics (networks statistics) while patterns most often correspond to particular subsets of entities (nodes and edges). Layers bring additional complexity to the computation of these metrics and patterns, as metrics and patterns may need to be computed across several layers. The visualization of the computed metrics and patterns needs to consider also these layers, thus, posing challenges to the data presentation. This working group analyzed the current network metrics and proposed novel metrics specifically for multilayer networks (see Section 4.4).
- Interaction (and Layer Creation): (see Section 4.2) This topic concentrated on interactive creation of layers in networks. While the input multilayer network may have predefined layers, in many use cases, the layers need to be adapted to the analytical task during network exploration. This working group has gathered requirements for interaction with layers, surveyed current solutions and their limitations. They have proposed novel approaches that will be pursued after the seminar.
- Visual Encodings The complex relationships between complex structures mean that traditional interactive visualizations need to be enhanced. Researchers from the various domains can exchange their ideas and thus start novel avenues in interactive visualization. The discussion of this working group focused on the visualization design – encodings. The group identified main requirements for visualization: aggregations, interactive layer editing, overview of all layers, details of an individual layer and exploration paths – top-down versus bottom up (see Section 4.5). These requirements are used to derive a design space of possible visualization approaches in future.
- Human Factors and Multilayer Networks This topic focused on the user’s point of view in the design of multilayer network visualization. This is a challenge as the complexity of multilayer networks results in a significant amount of cognitive load on the users. The group collected results from related work that can be used as guidelines for designing multilayer visualizations. It also identified gaps in literature for future research (see Section 4.1).
During the seminar, a number of sub-topics were identified that require further research: A unifying visualization framework, Novel Visual Encodings, Analytics and Attributes, Interaction, Evaluation, Use Cases, and Human Factors.
- A unifying visualization framework for multilayer networks: Currently, multilayer networks are referred to across communities using various names and concepts. A novel unified conceptual framework for multilayer network is needed that would be used for visualization, interaction and analytics purposes. It should extend the underlying mathematical framework  to meet the needs of the data and tasks associated with the various use cases, as well as existing visualization and interaction concepts.
- Novel visual encodings: The existing visualization techniques have limited scope for the broad range of data and tasks in the applications of multilayer networks. Therefore, novel visual encodings need to be researched that to enable data exploration across layers.
- Interaction: Visual exploration and analysis of multilayer networks requires novel interaction techniques that would allow to browse across layers and also to create new layers during the exploration process.
- Interdisciplinarity: The wide range of application domain problems sets novel problems that may be best addressed by new visualization approaches. The development of novel solutions for visual analysis of multilayer networks requires joined forces of application, visualization and analysis experts.
- Multiple layers and attributes: The complexity of multilayer networks often includes an additional dimension: The multivariate nature of node and edge attributes. This information needs to be encoded in the visualization and supported in analytical functions. This raises novel challenges.
- Network Analytics: Visual network analysis also covers the understanding the analytical relationship between layers (with respect to structure and/or attributes) and the layer comparison. The limitations of current analytical approaches and network metrics raises many interesting challenges and opportunities for developing new metrics for the multilayer use case.
- Evaluation & Human Factors: The human perspective on the complexity of the network structure and its visualization needs to be assessed. It covers a) the perceptual and cognitive aspects when interactively exploring the networks and b) a thorough empirical evaluation of the analytical paths and insights. The existing methodologies for such research should be adapted for the multilayer network case.
These topics will be discussed in the follow-up VIS 2019 Workshop "Challenges in Multilayer Network Visualization and Analysis". The workshop is co-organized by Dagstuhl Seminar organizers and participants: Fintan McGee, Tatiana von Landesberger, Daniel Archambault and Mohammad Ghoniem. The seminar will feature keynote, paper and poster sessions as well as discussion rounds on the above-mentioned topics.
- Mikko Kivelä and Alex Arenas and Marc Barthelemy and James P. Gleeson and Yamir Moreno and Mason A. Porter (1998). Multilayer networks. Journal of Complex Networks.
- Jan Aerts (Hasselt University - Diepenbeek, BE) [dblp]
- Daniel Archambault (Swansea University, GB) [dblp]
- Ariful Azad (Indiana University - Bloomington, US) [dblp]
- Kathrin Ballweg (TU Darmstadt, DE) [dblp]
- Fabian Beck (Universität Duisburg - Essen, DE) [dblp]
- Remco Chang (Tufts University - Medford, US) [dblp]
- Alberto Cottica (Edgeryders - Brüssel, BE) [dblp]
- Marten Düring (University of Luxembourg, LU) [dblp]
- Jean-Daniel Fekete (INRIA Saclay - Orsay, FR) [dblp]
- Mohammad Ghoniem (Luxembourg Inst. of Science & Technology, LU) [dblp]
- Martin Grandjean (Université de Lausanne, CH) [dblp]
- Jessie Kennedy (Edinburgh Napier University, GB) [dblp]
- Andreas Kerren (Linnaeus University - Växjö, SE) [dblp]
- Mikko Kivelä (Aalto University, FI) [dblp]
- Søren Knudsen (University of Calgary & University of Copenhagen, DK) [dblp]
- Stephen G. Kobourov (University of Arizona - Tucson, US) [dblp]
- Sylvain Legay (Luxembourg Inst. of Science & Technology, LU) [dblp]
- Matteo Magnani (Uppsala University, SE) [dblp]
- Maria Malek (EISTI - Cergy Pontoise, FR) [dblp]
- Mark McCann (University of Glasgow, GB) [dblp]
- Fintan McGee (Luxembourg Inst. of Science & Technology, LU) [dblp]
- Guy Melançon (University of Bordeaux, FR) [dblp]
- Anders Kristian Munk (Aalborg University, DK) [dblp]
- Bruno Pinaud (University of Bordeaux, FR) [dblp]
- Margit Pohl (TU Wien, AT) [dblp]
- Helen C. Purchase (University of Glasgow, GB) [dblp]
- Benjamin Renoust (Osaka University, JP) [dblp]
- Arnaud Sallaberry (University of Montpellier, FR) [dblp]
- Blaž Škrlj (Jozef Stefan Institute - Ljubljana, SI) [dblp]
- Christian Tominski (Universität Rostock, DE) [dblp]
- Paola Valdivia (INRIA Saclay - Orsay, FR) [dblp]
- Jason Vallet (University of Bordeaux, FR) [dblp]
- Tatiana von Landesberger (TU Darmstadt, DE) [dblp]
- Michael Wybrow (Monash University - Caulfield, AU) [dblp]
- Björn Zimmer (Univ. of Applied Sciences - Hagenberg, AT) [dblp]
- data structures / algorithms / complexity
- society / human-computer interaction
- Multilayer Network Visualization
- Complex Systems
- Graph Visualization
- Visual Analytics
- Social Network Analysis
- Biological Networks
- Geographic Networks