Dissertation defense (June 27, 2024): Érica Carneiro Queiroz da Silva

Abstract: With the advances of Machine Learning (ML), many statistical solutions were developed to solve Sentiment Analysis (SA) and Natural Language Processing (NLP) problems. However, until now, no statistical classification process would be able to resolve semantic and linguistic relationships in the same way as the human brain. According to the literature, only extremely robust tools would be able to generate correlations between the meanings of words, bringing the real symbolic meaning to texts. Ontologies are capable of representing semantic structures in order to correlate concepts in a specific domain, just as human beings associate words and receive messages according to their cultural, historical and social reality. Although recurrent in the area of Computer Science, ontologies are rarely applied to SA, with the exception of some examples in the English language, whose studies showed promising results for sentiment classification. Brazil currently does not have any robust framework that performs SA tasks with ontology-based knowledge base approaches and there are few studies using the methodology in Portuguese. This work presents a comparison of methods between lexical approaches and knowledge bases with ontological aspects, in addition we present a sentiment Portuguese – English parallel corpus. To that end, a review of different methods with ontology-based approaches applied to SA was initially carried out. Its primary objective was raising more discussions about possible expansion strategies for the field of SA in Brazil. We hope, thus, to awaken new reflections and a deeper look at digital humanities and their hybrid approaches, including ontologies, in the Luso-Brazilian scenario.