A Social Learning Model Based On Composite Strategy

With the development of social learning theory, social learning has become a hotspot for researchers. Social learning is generally studied based on the specific model,therefore different issues have different models. There are two typical social learningmodels, one is based on the beliefs of neighbors, the other is based on the Bayesianupdate rule. The social learning model based on beliefs of neighbors solvesthe problem of consistent convergence of belief simply and effectively, but it can’tguarantee the convergence to the true value. The social learning model based onBayesian Rule can achieve the value of true belief. But when we use this model, weneed to understand the topology of the network based on the model and large amountof computation. So it is subject to certain restrictions in the real application. Both ofthe models have some defects in practice, however, Ali Jadbabaie with others proposesa non-Bayesian model, in which the new belief is determined by the individual’s beliefand neighbors’ belief. This model can not only reduce the amount of computation, butalso to ensure convergence to the true value of belief, but the model requires that theindividual’s belief must exist and must be updated based on Bayesian rule, which isinconsistent with the actual situation in society.Taking into account the social network, the individual is not necessarily rational,not necessarily completely irrational, therefore, when we update individuals’ belief, wedon’t necessarily have adopted policies based on Bayesian updating rule, or notnecessarily based on the belief of neighbors, we present a composite strategy model. Itis more suitable for the real world, therefore, it is very important application value forthe study of complex policy model. The main work is as follows:1. In order to study social learning in complex social networks, a social learningmodel based on composite belief update strategy is proposed by considering theheterogeneity and complexity of the social individuals. At each time step, individualsin the model choose Bayesian update strategy or the update strategy based on theirneighbors’ beliefs according to the strategy selection probability. The simulation resultsshow that under some conditions such as the positive strategy selection probability, allthe social individuals can achieve asymptotic learning.2. We change the parameters in the model based on composite strategy to study how they affect individuals to reach asymptotic study and learning speed.