Recomendações de pontos de interesse baseadas no histórico e localizações de check-ins em redes sociais baseadas em localização.

Abstract

Location-Based Social Networks (LBSN) emerged with the purpose of allowing users to share, with their friends, information about points of interest (POIs) they visited. In this context, the ability to recommend new places for users to visit is important because it can eventually improve the overall user experience while using the system. The geographical context certainly influences the locations that the users choose to visit. Therefore, initially we analyzed this context separately, through the recommenders of POIs purely based on geographical information. Furthermore, we propose a new geographicaware recommender based on Gaussian Kernels. The results of our experiments demonstrated that the proposed model can achieve higher accuracy than the state-of-the art recommenders solely based on geographical information, in most of the cases evaluated. However, this same analysis showed that the geographical context alone is not able to generate recommendations owith high accuracy. So to model a new recommender of POIs, it is necessary to combine geographic information with other contexts in order to achieve high accuracy. Thus, we also propose a new recommender of POIs that can capture the preferences of users (similar to collaborative filtering techniques) and geographical information in a single model based on diffusion on graphs. This recommender aims to learn a personalized ranking of places to be recommended for each user taking into consideration the places visited by other users with similar preferences, the distances between the places visited and places candidates for recommendation, and the regions which the user visits more often. Our experiments showed that this model can be more efficient than state-of-the-art recommenders of POIs, also achieving an accuracy equal to or greater than the compared approaches. All experiments were conducted using real data from one of the most popular RSBL nowadays: Foursquare.