Community-based location inference in social media using supervised learning approach

Social media embed rich but noisy signals of physical locations of their users. Accurately inferring a user’s location can significantly improve the user’s experience on the social media and enable the development of new location-based applications. This paper adopts a supervised learning model—generalized additive model (GAM) to find the best community in a user’s online neighborhood to predict the user’s physical location. It proposes to use geographical proximity, structural proximity, and generic attribute metrics to characterize the goodness of the communities in the ego-net of a user and apply variable selection techniques to identify important community metrics for user location inference. Evaluating the effectiveness of GAM model with real social media data, we discover that GAM can choose better communities for location prediction than using an individual metric and GAM identifies median haversine distance, triangle participation ratio, and internal density as the top three significant metrics for community selection.