Point-of-Interest Recommendations: Capturing the Geographical Influence from Local Trajectories

Location-based social networks (LBSNs) provides rich information for Point-of-Interest (POI) recommendation. Previous researches tend to capture the geographical influence from the distance between every pair of locations visited by the same user, neglecting two key facts: (1) users usually visit most locations in a small area (we call local activity area); (2) not all locations are equally important: some locations are being visited multiple times, while others not. In order to better capturing those facts, we present the concept of local exploration and local trajectories. Based on this, we further propose the local trajectory movement model (LTMM), which determines user's movement probability via the local trajectories. We also propose the local activity similarity (LAS) to compute the similarity between users. Finally, we propose a fusion POI recommendation framework to combine LTMM and LAS. We evaluate the performance using two real-world data sets. The experimental results show that our work produces better recommendation accuracy than the popular POI recommendation algorithms.