Topic-based place semantics discovered from microblogging text messages

Location-based social network services (LBSNS) such as Foursquare are getting the highlight with the extensive spread of GPS-enabled mobile devices, and a large body of research has been conducted to devise methods for understanding and clustering places. However, in previous studies, the predefined set of semantic categories of places play a critical role in both discovery and evaluation of the results, despite its limited ability to represent the dynamics of the places. We explore beyond the predefined semantic categories of the places and discover topic-based place semantics through the use of Latent Dirichlet Allocation, by extracting topics from the text which people post on site. We also show the proposed method allows for understanding the temporal dynamics of the place semantics. The finding of this study is intended for, but not limited to, context aware services and place recommendation systems.