Mobile communications are increasingly contributing to the Internet traffic. Users move within cellular networks and generate traffics of different types. In this paper, we propose to categorize cellular base stations from multiple dimensions. Key social characteristics of the base stations are jointly considered, including traffic fluctuation, user non-determinacy, temporal homogeneity and usage diversity. Our categorization is based on spectral clustering, which is able to leverage the multiple social characteristics. The results of the categorization indicate different application environments and scenarios. We further establish social models with important parameters derived for each of the categories. The applications to energy efficient wireless networks are studied. The proposed model is of practical value to facilitate designing, simulating, and evaluating network deployment.