Catalogação de Objetos Educacionais Auxiliado por Aprendizado de Máquina para o Ambiente Virtual de Aprendizado do AlfaCon Concursos Públicos

Abstract

The research, conducted using data from the AlfaCon Concursos Públicos Virtual Learn ing Environment, aimed primarily to develop and optimize a classifier to assist in the cat aloging of its Educational Objects. The adopted method began with detailed exploratory research on two distinct data sets. To achieve the proposed objectives, various classifi cation algorithms were tested, ranging from traditional techniques to more contempo rary machine learning and deep learning approaches. Among the algorithms evaluated, the Rocchio, Boosting, Bagging, Naïve Bayes, K-Nearest Neighbors, Support Vector Ma chine, Decision Tree, Random Forest, Recurrent Neural Network, Convolutional Neural Network, Deep Neural Network, Recurrent Convolutional Neural Network, X-Class, and PECOS were explored. Additionally, there was a particular emphasis on the use of the Support Vector Machine, due to the algorithm’s performance aligned with non-functional requirements highlighted by the company. The study also benefited from the adaptation of previously established codes, building upon the seminal research of other academics in the field. However, the results presented challenges that need to be addressed for the use of classifiers to assist in the cataloging process. Among these, we mainly highlight the classifications made at different levels of the taxonomies that represent the organization of the contents of the disciplines studied by students and in which the Educational Ob jects should be cataloged. Furthermore, the research identified that the limited number of documents available for certain labels had a direct impact on the classifier’s accuracy. In conclusion, while the research provided valuable insights into the potential and limitations of various classification techniques in the AlfaCon environment, it also emphasized the need for further investigations and optimizations. Through a comprehensive approach, the study explored multiple techniques and methods, providing a solid foundation for improving the accuracy and robustness of classification in the specific context of AlfaCon Concursos Públicos and for understanding the challenges of using them in an applied environment.