Integrating social network diffusion into BDI-based simulations: application focus on large-scale evacuations

Computer simulations  of human behaviour are increasingly useful for understanding social phenomena. Often, different aspects of human behaviour are inspected using  different lenses: the social aspect is represented in diffusion models that capture  propagation of influences (e.g., information, opinions) through a social network; the cognitive aspect is  encoded in cognitive systems; and the physical aspect is encoded in agent based models. As these are intertwined in reality, modelling of all three aspects is useful in social simulations.

To this end, we propose a multi-level framework that maintains an agent's social, cognitive and physical aspects across distinct models, while collectively forming a consistent view of the conceptual agent. Our framework provides a generic approach of allowing diffusion processes to affect simulations via cognitive reasoning, and conversely allowing cognitive reasoning taking account of physical observations to impact diffusion processes. The framework is reusable, flexible and generic, and opens the way for the integration of a variety of existing social, cognitive and physical models.

We analysed through simulation experiments the extent to which the addition of information diffusion affects outcomes of an agent based simulation and observed that  simulation outcomes significantly vary when information diffusion is introduced. We also investigated how incorporation of cognitive reasoning affects outcomes of a diffusion process. Results revealed that diffusion outcomes can substantially differ when  reasoning on exogenous events is incorporated. Finally, we present an evacuation case study involving several thousand agents, where we demonstrate new modelling capability to support social and intelligent agents in evacuation simulations.  Our  framework sets the base for future research combining diffusion in social networks, cognitive reasoning and agent based simulation.