Avaliação de características para extração automática de aspectos

• Autor(a) principal: LIMA, Roberto Márcio Mota de
• Data de Publicação: 2018
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações da UFRPE
• Texto Completo: http://www.tede2.ufrpe.br:8080/tede2/handle/tede2/7855
• topic: Análise de dados; Extração de opinião; Modelo CRF; CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO

The increasing use of the Internet and other online interactions between people, such as chats, forum participation, e-commerce transactions, reviews of products and services, among others has led to the increasing need to extract, transform and analyze a vast amount of data, using a combination of text mining processes and others directly from the Web. Companies from several different sectors demand for customer feedback. Such institutions are increasingly interested in knowing what their real, or potential, customers say about them. From restaurants to hotels, from smartphones to cameras, the reviews are spread out over the internet, and they are essential to companies because they are created by people who somehow make, or use their services. People are more likely to express their opinions and practical experiences in products or services that they have used. Such feedback is important to business organizations and consumers. However, analyzing the hundreds or even thousands of customer input is difficult to be handled by humans. Therefore, it is necessary to provide concise information and concise summaries of such reviews. The Aspect-Based Sentiment Analysis is a recent trend and an approach that has much to explore since it has demonstrated relevant results in the literature as a more refined and punctual opinion extraction technique than those based purely on lexicon and rules. This MSc thesis aimed to research, implement and evaluate a new method for extraction of opinion terms in restaurant reviews taking into account and choosing some of the best features described in the state of the art. As the classification model, CRF was used, given its high efficiency as a conditional classifier. The results obtained showed good performance when compared to the principal related works of the area, highlighting the high coverage achieved by the developed method.