Pandemics, computer malware attacks, misinformation campaigns all of these phenomena have one
thing in common: some kind of information (a pathogen, a virus, fake news) that spreads across an
underlying structure of elements interconnected between each other by some kind of relationship
(physical contacts, public WiFi, social networks). In Computer Science, such structures are known as
graphs or networks. Within this formalization, the phenomena described above are referred to as
diffusion processes, and their dynamics have been studied to obtain models that allow researchers and
institutions to generally understand such complex events.
Visual Analytics (VA) is a discipline that combines automated techniques for analyzing large-scale data
with visualization, leveraging the cognitive capabilities of human perception to extract meaningful and
trustworthy insights. It is a plastic approach, that applies to countless application domains. Thanks
to the human-in-the-loop principle, the results of the analysis do not come from a black box (as in
some AI systems). Instead, the user has a direct and pivotal role in the whole process, generally
providing more reliable results than automated analysis alone.
Current methodologies to model diffusion processes over networks have used two approximations:
discretizing time and disregarding (or overly simplifying) the phenomenons inherent uncertainty.
Concerning the former, time is split into a sequence of evenly spaced instants (like the individual
frames of the movies). However, real-world phenomena happen at any time: using discrete time
means that we must define a time resolution by doing so, we lose the fine temporal details (i.e., the
exact sequence of events) that could be crucial, especially in decision making scenarios. The
probabilistic nature of the diffusion processes is reflected in an inherent uncertainty. How can it be
conveyed effectively and unambiguously to the user in the context of the analysis of a diffusion
process?
Event-based dynamic networks change in continuous time, overcoming time discretization. In the SANE
project, we aim to apply event-based network visualization and analysis to investigate complex and
uncertain diffusion phenomena. We strive to systematically characterize the topic in the visualization
research community, developing a common framework to foster research in the area. We will then
employ such framework to introduce and refine prototypes to analyze real data about diffusion
phenomena, improve current algorithmic solutions, and share lessons learned throughout the project
duration.