Uso de informações específicas de domínio em recomendações para turismo

Abstract

E-Tourism recommendation is the task of personalizing suggestion of destinations, points-of-interest and/or traveling routes to help tourists plan their trips. In this domain, most recommender systems rely on collaborative filtering information, and ignore side information associated with the items, such as text descriptions and distances between destinations. This dissertation aims at exploiting domain-specific side information in the context of destination recommender systems. Towards that goal, we propose a family of methods that takes into account domain-specific side information to recommend destinations (i.e., cities) to the user. More specifically, we consider two types of side information, namely textual reviews of attractions located in the target destination and geographic location. Experiments on a real-world dataset show that our best method greatly outperforms a number of relevant baselines with gains (in average precision) surpassing 10% in the top-10 recommendations.