A challenging problem is presented in this paper: recommending interesting places to tourists. The authors address the different challenges to provide recommendations that can be characterized by novelty and high relevance. To achieve that, they propose a Bayesian approach to extract the core social dimensions of tourists at different geographical locations after dividing people into various interest groups with high degrees of similarity. Different latent factorization techniques are designed to provide the best possible recommendations. The main techniques presented include non-negative matrix factorization and Bayesian probabilistic matrix factorization.

The main ideas of the paper are easy to grasp and highly relevant to several problems, since the issue boils down to providing recommendations for users. The paper builds on collaborative filtering techniques and further divides the users into groups to provide highly personalized recommendations. The techniques presented were tested with data collected from Twitter that refer to Foursquare check-ins. The authors argue for the importance of latent models to provide promising recommendations, and offer a comparison among different techniques to model the latent representations extracted.

The approach presented in this work can be useful in different domains, including disaster management and healthcare.