Emergent Dynamics in Multiplex Social Networks: Agent-Based Modeling of Information Diffusion for Misinformation Control

Information misrepresentation is widespread in multi-layered social networks which provide multiple avenues to communicate information. As such, it presents significant opportunities for both information integrity and public discourse to be undermined by disinformation. This paper outlines a new agent-based model, developed to capture emergent dynamics of multi-layered social networks and to help identify technical means to mitigate information misrepresentation in complex systems. A key component of this research includes a novel Multi-Layer Information Diffusion Model (MLIDM), integrating both cross-layer communication among agents, as well as heterogeneous agent behaviors and adaptive intervention strategies. Our methods employ a three-stage process to model misinformation spreading in multi-layered social networks: (1) constructing the layered architecture of a multi-layered social network; (2) using agent-based modeling techniques to simulate the behavior of each layer and to update beliefs based upon those behaviors; and (3) developing targeted intervention strategies that utilize the topology of the network and patterns of information flow. Through experimental testing of both synthetic and actual data sets (e.g., Twitter COVID-19 dataset, n = 50,000 tweets), we demonstrate that our method can reduce the rate at which misinformation spreads by 43.7% when compared to a baseline model. Furthermore, we find that our method improves early detection of misinformation by 28.5%. In addition to these quantitative results, we also find evidence for critical thresholds for the timing of interventions, where interventions applied during the first 15% of the diffusion timeline are significantly more effective than interventions applied later in time (E = 0.67 vs. E = 0.21, Cohen's d = 1.82). Finally, we find evidence of phase transitions in 23% of runs and spontaneous polarization in 67% of runs, indicating emergent phenomena resulting from both agent-network interactions. Therefore, our results will aid in the theoretical development of information dynamics in complex systems, while providing practical guidance for the design of effective counter-measures to misinformation in digital environments.