Optimal combination policies for adaptive social learning