Point of Interest Recommendation using Implicit Trust based on Combining Ratings and Check-ins of Smartphone Users

This paper introduces a hybrid model called Implicit Trust based on Combining point-of-interest Ratings and user Check-ins (ITCRC) to address the cold-start challenges commonly associated with trust-based collaborative filtering methods. The model combines Point of Interest (POI) ratings and user check-ins to estimate implicit trust, facilitating location recommendations in a Location-Based Social Network (LBSN). In the Yelp dataset, the ITCRC model's trust and prediction matrices are calculated using Trust based on Ratings (TR), Trust derived from Check-ins (TC), and Trust based on the Hybridization of ratings and check-ins (TH), as well as three approaches derived by adapting O'Donovan's trust formula to the LBSN context. These six approaches are then compared using sparsity metrics and evaluation parameters such as RMSE, precision, and recall. The comparisons revealed that the TH approach significantly reduces the data sparsity of the prediction matrix by 36.08%, the TR and TC approaches improve the relevance of the recommendations (0.77% of precision and 0.99% of recall), and the OR, OC, and OH approaches improve the prediction accuracy by 0.2% in terms of RMSE.