We propose an influence-based aggregative learning framework that facilitates the emergence of social norms in complex networks and investigate how a norm converges by learning through iterated local interactions in a coordination game. In society, humans decide to coordinate their behavior not only by exchanging information but also on the basis of norms that are often individually derived from interactions without a centralized authority. Coordination using norms has received much attention in studies of multi-agent systems. In addition, because agents often work as delegates of humans, they should have 'mental' models about how to interact with others and incorporate differences of opinion. Because norms make sense only when all or most agents have the same one and they can expect that others will follow, it is important to investigate the mechanism of norm emergence through learning with local and individual interactions in agent society. Our method of norm learning borrows from the opinion aggregation process while taking into account the influence of local opinions in tightly coordinated human communities. We conducted experiments showing how our learning framework facilitates propagation of norms in a number of complex agent networks.