Information Sources Driving Social Influences: A New Model for Belief Learning in Social Networks

Non-Bayesian models for learning in social networks, such as the DeGroot model, are focused on updating beliefs using social influence weights, and study the achievement of convergence of beliefs to a consensus. In this paper, we propose a new construct to capture the notion of agents using additional information sources, such as media, to obtain multiple affirmations of belief information through an information score. We use this new construct to create a feedback learning loop that allows agents to learn beliefs with social influences varying dynamically with the credibility of the information on such beliefs. We build a new social learning mechanism, without the constraining row-stochasticity assumptions on social influence, and show that with information sources driving social influences, the beliefs can converge, and not necessarily to a consensus, even with strong connectivity. In this process of learning agents with a similar alignment between exogenous information sources and beliefs ´group´ together, and this reflects in the structure of social influences. The results of the new learning algorithm are demonstrated on a small, fully connected social network.