Social learning in multi-true-state networks

Social learning focuses on the opinion dynamics in the society, which has attracted more and more researchers recently. Different from the existing results on the consensus of social learning in complex networks with one true state, in this paper we study the social learning in the network society with multi-true-states. A new network social learning model is constructed, where agents from different groups receive different signal sequences generated by different true states. Each agent updates his belief by combining a Bayesian rule on the external signal and a non-Bayesian rule related with his neighboring agents. We analyze the dynamical process, and find that the beliefs of all agents are oscillating all the time and can not access to their corresponding true states, which is totally different from the consensus on social learning with one-true-state. Furthermore, by calculating the largest Lyapunov exponents, chaos is found in the social learning with multi-true-sates.