Providing location recommendations has become an essential feature for location-based social networks (LBSNs) because it helps users explore new places and makes LBSNs more prevalent to users. Existing studies mostly focus on introducing new features that affect users' check-in behaviours in LBSNs. Despite the difference in the type of the features exploited, they mostly follow the same principle - characterizing dependencies between the probability of a user visiting a points-of-interest (POI) and each feature separately. However, the decision of a user on where to go in an LBSN is driven by multiple features that act simultaneously. On the other hand, applying a full model which considers all the features jointly suffers from over fitting, as for each user there is limited available data. We propose an intermediate solution by fragmenting the model into multiple partial models which each takes the subset of the features as the input. The proposed approach focuses on building the personalized partial models which then combine them by applying an additive approach. We further introduce a partition-based approach to identify the hidden patterns from the geographically clustered check-in data. Experiments on two datasets from Foursquare show that our proposed method outperforms the state-of-the-art approaches on POI recommendation.
- location-based social network (LBSN)
- point-of-interest (POI)