From Foursquare to My Square: Learning Check-in Behavior from Multiple Sources

Abstract

Location-based services often use only a single mobility data source, which typically will be scarce for any new user when the system starts out. We propose a transfer learning method to characterize the temporal distribution of places of individuals by using an external, additional, large-scale check-in data set such as Foursquare data. The method is applied to the next place prediction problem, and we show that the incorporation of additional data through the proposed method improves the prediction accuracy when there is a limited amount of prior data.