Social learning with bounded confidence and probabilistic neighbors

This paper investigates a social learning model, in which each individuals´ neighbors are composed by two parts: one consists of those who have similar beliefs; the other consists of those picked up according to certain probabilities. Each individual updates her beliefs by combining the Bayesian posterior beliefs based on her private signals and weighted averages of the beliefs of her neighbors. It is shown that the whole group may be divided into clusters by only communicating with those having similar beliefs, especially when bound of confidence is relatively small. Adding probabilistic neighbors guarantees the whole group achieving consensus effectively, which means communicating beyond the difference of beliefs may improve social learning.