O turismo é uma área que teve grandes impactos com expansão da internet. Hoje é possível planejar uma viagem de casa, usando somente informações da web. No entanto, os usuários chegaram a um ponto em que a quantidade de dados fornecidos pode ser mais confusa do que esclarecedora, causando um problema chamado de sobrecarga de informações. Assim, este trabalho se concentra reduzir este problema. Para isso, utiliza técnicas de mineração de texto para sumarizar as opiniões dos usuários e para entender o quão próximo ela está da classificação discreta dada por eles para um lugar ou atração no TripAdvisor. Como estudo de caso escolhemos a cidade de Tiradentes, localizada no interior de Minas Gerais, Brasil.

The amount of available data on the web has grown exponentially, mostly due to the emergence of the Collaborative Internet, in mid-2006, which turns the process of obtaining information into a hard task. This way, several computational techniques have been used in order to automate the exploitation and analysis of data, such as Text Mining techniques, Topic Modeling (TM), which establishes relationships between text documents and discussion topics through the present words, and Sentiment Analysis (SA), whose objective is to identify sentences' polarity; Complex Networks modeling, which seek to capture the dynamics of complex systems, present in social networks; and Recommendation Systems, which assist with decision making and whose operation resides in the suggestion of items that have not yet been evaluated by a user, such as traveling to a new place or trying another meal from a menu. The Tourism scenario is also included in the context of massive data generation and advances in techniques to deal with them. In this case, specialized travel platforms, like Tripadvisor, have a major role since they concentrate a large amount of data about users and their experience in Points-of-Interest (POI). Therefore, this work proposes a new approach to a predictive model for POI recommendation systems based on the construction of a Complex Network and the use of specific techniques for its structural analysis. The city chosen to validate these objectives was the city of Tiradentes, Minas Gerais, whose geographic proximity and tourism-oriented economy make it a good choice. The results obtained show that a predictive model based on Complex Networks does not overcome the error obtained by baseline algorithms, however, it brings a good ranking correlation between what was predicted and the real result, which makes it a good option for recommendation systems.