Improving location prediction using a social historical model with strict recency context

Location-based Social Networks (LBSN) are a popular form of social media where users are able to check-in to locations they have visited and share these check-ins with their friends. An important problem in LBSNs is the prediction of a user's next check-in location, for purposes such as showing location-specific recommendations or advertisements. One state-of-the-art algorithm is the Social Historical Model (SHM) which utilizes a user's historical check- ins and social links to predict his/her next check-in location. Observing various LBSN user characteristics, we improve the SHM by exploiting the recency effect (where users are more likely to revisit places from their recent past than the more distant past) and place-links (where two friends share a common temporal check-in). Using two Foursquare datasets, we then demonstrate how these modifications improve the overall location prediction accuracy of the SHM.