Hierarchical geographical modeling of user locations from social media posts

With the availability of cheap location sensors, geotagging of messages in online social networks is proliferating. For instance, Twitter, Facebook, Foursquare, and Google+ provide these services both explicitly by letting users choose their location or implicitly via a sensor. This paper presents an integrated generative model of location and message content. That is, we provide a model for combining distributions over locations, topics, and over user characteristics, both in terms of location and in terms of their content preferences. Unlike previous work which modeled data in a flat pre-defined representation, our model automatically infers both the hierarchical structure over content and over the size and position of geographical locations. This affords significantly higher accuracy --- location uncertainty is reduced by 40% relative to the best previous results [21] achieved on location estimation from Tweets.

We achieve this goal by proposing a new statistical model, the nested Chinese Restaurant Franchise (nCRF), a hierarchical model of tree distributions. Much statistical structure is shared between users. That said, each user has his own distribution over interests and places. The use of the nCRF allows us to capture the following effects: (1) We provide a topic model for Tweets; (2) We obtain location specific topics; (3) We infer a latent distribution of locations; (4) We provide a joint hierarchical model of topics and locations; (5) We infer personalized preferences over topics and locations within the above model. In doing so, we are both able to obtain accurate estimates of the location of a user based on his tweets and to obtain a detailed estimate of a geographical language model.