Population Distribution Projection by Modeling Geo Homophily in Online Social Networks

Abstract

Today, many applications depend on the projection on the population distribution in geographical regions, such as launching marketing campaigns and enhancing the public safety in certain densely-populated areas. Demographic and sociological researches have provided various ways of collecting people's trajectory data through offline means. However, collecting offline data consumes a lot of resources, and the data availability is usually limited. Fortunately, the wide spread of online social network (OSN) applications over mobile devices reflect many geographical information, where we could devise a light weight approach of conducting the study on the projection of the population distribution using the online data. In this paper, we propose a geo-homophily model in OSNs to help project the population distribution in a given division of geographical regions. We establish a three-layer theoretic framework: It first describes the relationship between the online message diffusion among friends in the OSN and the offline population distribution over a given division of regions via a Dirichlet process, and then projects the floating population across the regions. Evaluations over large-scale OSN datasets show that the proposed prediction models can characterize the process of the formation of the population distribution and the changes of the floating population over time with a high prediction accuracy.