The social behavior in cognitive radio networks is studied using analysis tools in social networks. A recommendation system is proposed for cognitive radio, thus incurring the channel preference propagation in the corresponding random geometric network. A mean field based ordinary differential equation is used to describe the dynamics of the channel preference propagation in cognitive radio networks. The conditional distribution of random geometric graph is studied. The convergence and the steady state of the mean field equation are discussed. Numerical simulations are used to demonstrate the properties uncovered by the analysis.