A Hierarchical Decision-Making Framework in the Network Environment with Social Learning and Forgetting

Abstract

Modeling and analysis of human behaviors in social networks are essential in fields such as online business, marketing, and finance. However, the establishment of a generalized decision-making framework for human behavior is challenging due to different decision structures among individuals. This research proposes a new hierarchical human decision-making framework based on the evolution of preferences on alternatives over time. To this end, a well-known cognitive psychological model, Decision Field Theory (DFT) which is one of dynamic human decision-making models based on the evolution of preferences on the options over time, is utilized and extended to represent human forgetting and learning procedures with the properties of memory loss experience and influences under social interactions. The equilibrium status of social networks within this framework is derived as an explicit formula under the independent and identically distributed (IID) conditions on weight values, which facilitates the identification of limiting expected and covariance matrices for preference values. The extension establishes a hierarchical human behavior model in social networks by incorporating the dynamics of top-down and bottom-up information flows, which enables the better understanding of different behaviors in social networks such as innovation diffusion and opinion formation. The validity of the proposed model is demonstrated via agent-based simulation under various scenarios. In particular, simulation is used to analyze the impact of network structures (e.g., random, small-world, ring-lattice, and scale-free) as well as the significance of inherent society characteristics (e.g., conservative, neutral, and progressive) on the equilibrium states. The findings confirm that the diffusion process within the proposed model propagates fastest in the random network and slowest in the ring-lattice network. It is also shown that interaction among people affects the agent’s decision within the proposed models and intensifies the embedded society characteristics, which helps to analyze irregular behaviors such as information cascades in social networks. Two major applications of the proposed models in this dissertation are 1) disaster management with social sensing and 2) real-time border surveillance. The simulation results reveal that the proposed models allow for better disaster management strategies in natural disasters by increasing the efficiency of prepositioning supplies and by enhancing the effectiveness of disaster relief efforts. Moreover, physics-based simulation developed in the Unity3D engine has a potential to increase the modeling accuracy of a border surveillance system by enhancing the estimation of drug-traffickers’ behaviors with real-time environmental information, which will, in turn, help establish an effective control system in border areas.