Learning analytics e avaliações online: uma metodologia orientada a ciência de dados em grafos

• Ano de defesa: 2023
• Autor(a) principal: Melo, Elvis Medeiros de
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Rio Grande do Norte; Brasil; UFRN; PROGRAMA DE PÓS-GRADUAÇÃO EM ENGENHARIA ELÉTRICA E DE COMPUTAÇÃO
• Palavras-chave em Português: Learning analytics; Ciência de dados em grafos; Grafos; Avaliação online; Machine learning; eXplainable artificial intelligence; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA
• Link de acesso: https://repositorio.ufrn.br/handle/123456789/51948

Evaluation is an important tool to correct failures and promote successes in learning processes. It is one of the pillars of education, along with the curriculum and the teaching and learning process. With the pandemic of COVID-19, the Federal University of Rio Grande do Norte (UFRN) has adopted the Multiprova platform to support online assessment. With the computerization of assessment processes comes the need to use Learning Analytics (LA) and graph data science techniques to understand student behavior and identify student profiles from online assessment resolution records. When analyzing the literature, there is a need to transform interpretable models into education. For this, data visualization techniques and eXplainable Artificial Intelligence (XAI) need to be considered. The proposed hypothesis is that it is possible to use this data to gain insights into the learning process and student profiles using LA techniques such as graph modeling and Machine Learning (ML). To this end, the theoretical framework is presented about the topics that make up the object of study, such as graphs, LA, and online evaluations. Among the results, a systematic literature review pointed to 40 papers involving LA and online assessments, but no papers used graph metrics with LA techniques such as ML to analyze student performance. In addition, two case studies were modeled according to the proposed graph data science oriented methodology, with about 78.75% of the graph features explaining the ML models used. We see the importance of using graph features in LA techniques in identifying insights about student learning considering their journey in online evaluation, as well as using graph metrics and XAI for the interpretation of the results.