What is this place? Inferring place categories through user patterns identification in geo-tagged tweets

Abstract

Online social networks such as Facebook and Twitter have started allowing users to tag their posts with geographical coordinates collected through the GPS interface of users smartphones. While this information is quite useful and already indicative of user behavior, it also lacks some semantics about the type of place the user is (e.g., restaurant, museum, school) which would allow a better understanding of users' patterns. While some location based online social network services (e.g., Foursquare) allow users to tag the places they visit, this is not an automated process but one which requires the user help. In this paper we exploit the dynamics of human activity to associate categories to GPS coordinates of social network posts. We have collected geo-tagged tweets of a large city through Twitter. A supervised learning framework takes the tweets spatial-temporal features and determines human dynamics which we use to infer the place category. Our results over the data show that the prediction framework is able to accurately identify if a place is of a certain category given its user activity patterns. The average accuracy is about 70%, reaching the highest accuracy for work (90%) and educational places (80%). Moreover the framework identifies the category of a place, with an accuracy up to 66%, finding out where people eat and drink, go for entertainment, or work/study.