Location Prediction via Social Contents and Behaviors: Location-Aware Behavioral LDA

Massive user generated contents in social media platforms like Twitter asks for analyzing conversations on topics of interest. Unlike traditional text-dominant settings, such platforms are rich in social interactions that are accompanied with textual content, e.g., reply and mention. Furthermore, the proliferation of mobile devices equipped with geo-positioning components creates a growing volume of social content with geographical locations. However, social content analysis considering location has not been well studied. It is still vague whether a linkage between location and topics from social interactions exists. In this paper, we propose a generative model, called Location-aware Behavioral LDA (La-LDA), that is not only addressing what are topics of interest in social content, but also linking topics with 1) user interactions, and 2) locations. We analyzed Twitter and our experimental studies show that our model can find location-aware topics relevant to user behaviors that are specified by the social media platform and also can be applied for location prediction.