Incremental tree-based successive POI recommendation in location-based social networks

This paper proposes a sequential rules-based recommendation system, called STS-Rec. It addresses the main drawbacks of sequential patterns mining approaches for POI (Point of interest) recommendation by considering both temporal and social influences to perform short-term recommendations. STS-Rec first transforms mobility data into location sequences. Then, it incrementally mines sequential recommendation rules in these sequences. In contrast with standard sequential recommenders, the proposal (1) discovers rules that tolerate locations' order variations by loosening the strict ordering constraint of location sequences, (2) builds a tree-based model to incrementally mine recommendation rules, and (3) supports short and long-term POI recommendation by using a user-defined window by extracting patterns that appear within a maximum number of consecutive locations. To take the temporal influence into account, STS-Rec adapts its mining strategy to include the temporal context in location data. Hence, the conventional rule mining problem is redefined to mine time-extended recommendation rules. An experimental evaluation conducted on two large-scale real check-in datasets from Gowalla and Brightkite shows that the proposed model outperforms two state-of-the-art sequential models in terms of accuracy and coverage.