Point-of-Interest Recommendations via a Supervised Random Walk Algorithm

Recently, location-based social networks (LBSNs) such as Foursquare and Whrrl have emerged as a new application for users to establish personal social networks and review various points of interest (POIs), triggering a new recommendation service aimed at helping users locate more preferred POIs. Although users' check-in activities could be explicitly considered as user ratings, in turn being utilized directly for collaborative filtering-based recommendations, such solutions don't differentiate the sentiment of reviews accompanying check-ins, resulting in unsatisfactory recommendations. This article proposes a new POI recommendation framework by simultaneously incorporating user check-ins and reviews along with side information into a tripartite graph and predicting personalized POI recommendations via a sentiment-supervised random walk algorithm. The experiments conducted on real data demonstrate the superiority of this approach in comparison with state-of-the-art techniques.