Methods for identifying aspects in opinion texts in Portuguese: the case of implicit aspects and their typological analysis

• Ano de defesa: 2023
• Autor(a) principal: Machado, Mateus Tarcinalli
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Aspectos explícitos; Aspectos implícitos; Explicit aspects; Identificação de aspectos; Identification of aspects; Implicit aspects; Mineração de opinião; Opinion mining
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/55/55134/tde-16012024-151720/

This work has as object of study the methods for identifying aspects, which are applications derived from the aspect-based sentiment analysis, and the area of natural language processing. The aspect-based sentiment analysis is focused on analyzing opinionated texts (texts containing opinions) seeking to identify and relate feelings and aspects (characteristics) of a given entity (products, services, among others). This work is focused on the first stage, which is the identification of aspects, emphasizing the so-called implicit aspects, that is, those that are not explicitly mentioned in the texts. In the course of this work, we implemented, adapted, and improved several aspect extraction methods, including frequency-based, rule-based, hybrid, machine learning, pre-trained language models, and large language models. We also developed novel resources such as a corpus extension with annotation of implicit aspects, lexicons of nominal forms, and a typology of implicit aspects. The latter allowed a better understanding of the knowledge necessary to relate an implicit aspect clue with an aspect and its use allowed a clear view of the strengths and weaknesses of each implemented method in relation to the detection of implicit aspects.