Improving recommendation accuracy for travelers by exploiting POI correlations

Personalized point-of-interest (POI) recommendation is a challenging task in location-based-service (LBS). Previous efforts on POI recommendation mainly focus on local users. According to user’s activity areas, e.g., home and workplace, nearby locations have higher probability to be recommended. However, in many practical scenarios such as urban tourism, target users are usually out-of-town travelers. Their preferences are hard to model due to sparse distributed check-ins. In this paper, we manage to improve the location recommendation accuracy for travelers, via finding correlations between different POIs. For cross-city POIs, the influence of travel intent (I), e.g., business trip and family trip, is studied. For local POIs, we focus on their geographical neighbors (N). In addition, reviews (R) are introduced to bridge the gap between distant POIs and make recommendation explainable. Incorporating these three factors into the learning of latent space, a novel matrix factorization approach (INRMF) is proposed. Further experiments conducted on real dataset show our approach is competitive against state-of-art works.