Implicit Trust-Based POI Recommendations Using User Check-Ins

Location-based social networks (LBSNs) offer users new opportunities to explore Points Of Interest (POIs). Collaborative filtering-based recommendation systems use implicit trust derived from ratings to predict future visits but suffer from biases and user inactivity. To overcome these limitations, some studies incorporate trust derived from friendships or shared visits. This article proposes an alternative approach that leverages user check-ins to infer implicit trust and improve recommendation accuracy. By relying on check-in behavior similarity, our method (TC-new) enables better prediction of future POI visits. Experimental results, based on the Yelp dataset, show that TC-new outperforms traditional methods, including Matrix Factorization and O'Donovan's approach, by offering a better balance between RMSE, MAE, PRECISION, and RECALL. While O'Donovan's approach excels in PRECISION and coverage, TC-new stands out for its relevance and reliability. Thus, future research on hybrid approaches could help overcome the challenges related to the cold-start problem and data sparsity.