Improving recommendations with implicit trust propagation from ratings and check-ins

Abstract

This paper investigates how the propagation of implicit trust between users affects the quality of point-of-interest (POI) recommendations in location-based social networks (LBSNs). Through the analysis of user interactions via ratings and check-ins, this work proposes a recommendation model known as propagation of rating/check-in for implicit trust (PRCT). This model relies on two primary approaches: Similarity trust rating (STR), which utilizes user ratings, and similarity trust check-in (STC), which focuses on check-ins data. Both approaches employ trust propagation to enhance their similarity matrices between users. An evaluation of the PRCT model using the Yelp dataset shows that the STR approach surpasses other variants in terms of PRECISION and RECALL, while the STC approach demonstrates superior performance in terms of RMSE. Furthermore, while trust propagation in the PRCT model increases the density of its similarity matrices, it does not consistently enhance its PRECISION parameter. Only the similarity Jaccard check-in (SJC) and similarity cosine check-in (SCC) approaches show a significant improvement of this parameter.