Community-aware empathetic social choice for social network group decision making

Abstract

To study how interdependent and self-interested decision makers collectively make a choice in highly complex social environments, this paper presents a novel and powerful Community-Aware Empathetic Social Choice (CAESC) model, which considers decision makers in social networks as autonomy-oriented agents, who derive utility based on both their own intrinsic preferences, and empathetic preferences determined by the satisfaction of their intra-community neighbors. We devise optimization approaches for identifying community structures with local (global) social welfare maximizing under two varieties of CAESC models: The local CAESC model first identifies the non-cooperative behaviors by calculating the structure-preference coordination coefficients between pairwise decision makers, then employs the Logit response like dynamics process in a carefully defined potential game to find the optimal community structure that maximizes the local social welfare. To find the optimal community structure that maximizes the global social welfare, the global CAESC model is composed of three coupled phases during each stage: (i) opinion evolution within each community using the classic Jacobi method; (ii) structure-preference coordination matrix updating based on the current opinion vector; and (iii) local optimal community structure updating based on the local CAESC model. Extensive experiments on both randomly generated and real-world social networks with synthetic and real-world preferences confirm that neglecting the empathetic effect and community structure usually yields sub-optimal group decisions which degrade the social welfare of network members. Our experiments also show that, the higher societal empathy, the rougher preference distribution and the denser network structure will facilitate the maximal social welfare reaching in both local and global CAESC models; meanwhile, by comparing with the typical empathetic decision making models and some recent proposed consensus reaching methods, our proposed approaches show relatively significant advantages in terms of candidate ranking and decision efficiency.