Organização da informação: um modelo semiautomático de classificação de atrações em perfis turísticos usando aprendizado de máquina

Abstract

The technological evolution paradigm has brought a disruptive change people's behavior, who now make decisions based on the content they consume on the Internet. This aspect is no different in the Tourism industry, where new technologies and the sharing of reviews allow users to seek information to support decisions such as choosing a destination, accommodation, attractions, food, among others. Reviews provide an important source of information; however, their volume can make it difficult to extract knowledge and use it effectively. How to find out if a particular point of interest with more than 100,000 opinions written in unstructured text is similar to what a tourist is looking for? This question motivates the development of this research, which has as its direct objective the creation of a model that allows transforming the reviews made by users into tourist classes (profiles). In the literature, some works try to address the problem of point of interest classification using reviews. In addition, the use of profiles in tourism is common, as a way of classifying destinations and tourists. In this sense, this study can present an additional view on both aspects, while allowing the joining of tourist profiles with review's information. The work presents an applied research, based on Pragmatism, of a hybrid nature with an exploratory objective. It uses the reviews organization as they quality nature as a source for a quantitative exploration analysis. The methodology presents the creation and validation of a classification model at three levels. At the Conceptual Level, knowledge is explored from domain experts, such as the creation of a set of 12 tourist profiles and definition of destinations to be used in the research. At this level, 3.4 million tourist reviews written in Portuguese are also collected. At the Technological Level, information is organized, represented and an automatic text classification process is carried out using different Machine Learning techniques. The Validation Level presents a comparison between automatic methods and a classification carried out by specialists. The best performing method is used to explore compatibility between destinations, attractions, states, countries and profiles, as well as the differences between the popularity and similarity of destinations with a profile. It also explores the similarity between destinations and the profile variation of the most visited destinations. The specific results present interesting discoveries in tourism, such as the identification of the best destinations for each profile, the most popular destinations that are not the most relevant for a profile or the identification of a very high degree of similarity between national and international destinations. The model performance above 70% accuracy, using technology and specialists offer an important alternative for models of knowledge organization, mainly due to the dynamism and exponential growth of content on the Internet. The results can help tourists looking for certain experiences, governments to promote tourism for a specific audience or private companies that aim to offer targeted products and services. Regardless of the actor in the process, the organization and classification of tourist information turn the decision-making process easier.