Framework para mineração de opiniões em mídias sociais para descoberta de conhecimento do cliente

• Autor(a) principal: Batista, Huoston Rodrigues
• Data de Publicação: 2017
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações da Uninove
• Texto Completo: http://bibliotecatede.uninove.br/handle/tede/3046
• topic: mineração de dados; mineração de textos; mineração de opiniões; conhecimento do cliente; redes sociais; data mining; text mining; opinion mining; customer knowledge; social networks; CIENCIA DA COMPUTACAO::SISTEMAS DE COMPUTACAO

With the spread of the Internet and popularization of mobile technologies, relations between customers and businesses have been transformed. Comments about the company, products or services, previously restricted to circles of friendship, now are shared consistently and prolifically on social networks and websites specializing in receiving opinions from customers regarding their experiences. This phenomenon provides opportunities for knowledge discovery from these opinions, but also challenges, considering that, given their nature and form, customer reviews consist of unstructured data, which in turn require specific treatments. This research aims to present a opinion mining framework for customer knowledge discovery in relation to their experiences in restaurants, based on unstructured data extracted from social networks, applicable to the reality of Small and Medium Enterprises. The social network chosen for the development of this research was TripAdvisor, from which data were extracted from four restaurants through the technique of web scraping. The data of the first company were used to develop and refine the framework, which in turn, was applied to the data of the other companies. The data were processed through a series of text mining techniques, including Sentiment Analysis and Topic Modeling using the tidy data approach, such as tokenization, normalization, removal of stop words, removal of special characters and numbers, creation of bi-grams, calculation of relevance of terms, comparisons and counts. As main results, we highlight the generation of summaries and graphic visualizations that contributed to evidence knowledge about the relations between several expressions and terms that were not obvious. These, in turn, were discovered from the analysis made, which allowed finding latent relationships between terms cited by different customers. The Sentiment Analysis allied to the Topic Modeling revealed that the aspects most addressed by the clients refer to the food, the place, and the service, varying in intensity and polarity. The practical contribution of this work lies in the application of Text Mining to reveal patterns and enable the discovery of knowledge from the opinions of customers extracted from social networks. The framework proposed and applied in this research proved useful as a tool to better understand the client, his expectations, and even his frustrations, thus generating knowledge about the clients for the benefit of the company.