Social media is pervasive throughout today's society.  Reports of, and reactions to, events can propagate locally, nationally, and internationally almost instantly, and can themselves trigger a cascade of reactions and events.  Within this reality, we believe that it is important for any large-scale simulation of an engaged population using human behaviour modelling to incorporate social media.  This paper reports on the design, implementation, and integration of the social media component within a broader simulation of Greater London with application to decision support systems, the training of decision makers, and course of action analysis.  Within the social media component, live social media messages were combined with synthetically generated messages and presented in real-time, while analytics aggregating sentiment expressed in the messages were displayed to decision makers throughout the course of the simulation.  The synthetic messages were generated by a trained AI model; each was related to the different emotions of happiness, sadness, fear, anger, joy, surprise and assigned to a member of the simulated population.  The integration of an externally triggered event in the simulation, a power outage, resulted in a change in behaviour of the simulated population and consequently a change in the resulting tone and emotion reflected in the synthetic social media messages.  Finally, initial user feedback is reported, and considerations for additional factors to influence the synthetically generated social media content based on cognitive state, demographic attributes, and extensions to the AI models are discussed.

Keywords
AI,DECISION,HUMAN PERFORMANCE

Additional Keywords
Social Media