Geo-Social Clustering of Places from Check-in Data

In this paper, we develop an algorithm to cluster places not only based on their locations but also their semantics. Specifically, two places are considered similar if they are spatially close and visited by people of similar communities. With the explosion in the availability of location-tracking technologies, it has become easy to track locations and movements of users through user "check-ins". These check-ins provide insights into the community structure of people visiting a place, which is leveraged and integrated into the proposed geo-social clustering framework called GeoScop. While community detection is typically done on social networks, in our problem, we lack any network data. Rather, two people belong to the same community if they visit similar geo-social clusters. We tackle this chicken-and-egg problem through an iterative procedure of expectation maximization and DBSCAN. Extensive experiments on real check-in data demonstrate that GeoScop mines semantically meaningful clusters that cannot be found by using any of the existing clustering techniques. Furthermore, GeoScop is up to 6 times more pure in social quality than the state-of-the-art technique. The executables for the tool are available at http://www.cse.iitm.ac.in/ ~simsayan/software.html.