Estudo empírico de avaliação inteligente computadorizada de questões abertas baseado em colaboração

• Ano de defesa: 2009
• Autor(a) principal: Corrêa, Kleber Pullig
• Orientador(a): Machado, Paulo César Miranda
• Banca de defesa: Machado, Paulo César Miranda, Nalini, Lauro Eugênio Guimarães, Martins, Weber, Lemos, Rodrigo Pinto, Fleury, Claudio Afonso
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Goiás
• Programa de Pós-Graduação: Programa de Pós-graduação em Engenharia Elétrica e da Computação (EMC)
• Departamento: Escola de Engenharia Elétrica, Mecânica e de Computação - EMC (RMG)
• País: Brasil
• Palavras-chave em Português: Avaliação computadorizada de questões abertas; Inteligência artificial; Correção colaborativa
• Palavras-chave em Inglês: Computere‟s evaluation of open questions; Artificial intelligence; Collaborative correction
• Área do conhecimento CNPq: OUTROS
• Link de acesso: http://repositorio.bc.ufg.br/tede/handle/tede/12529

This research submits to experimental tests in real-world environment, the Conexionist Collaborative Intelligent Evaluation System idealized by Martins (2004) and tested with simulations by Guimarães (2004), who have noticed consistent and promising results when dealing with synthetic datas. The system explores the Learning Theory by Kolb (1984), Peer Collaboration Evaluation and Artificial Intelligence (multilayer perceptrons neural networks), which are capable to apply and correct tests with open (free) questions. To get a real-world data, some students from the course of Engineer registered at Universidade Federal de Goiás (UFG) on the period of june 2008 to july 2009 were taken in heterogeneous groups composed by men and women in age from 16 to 26 years old and at least 18 students per test. The software SAICOweb, developed specifically to this data collection, allows the computerized evaluations of open (free) questions, by using methods and stages described at Guimarães (2004) research. SAICOweb system were experimental validited in a satisfactory way by an unique group composed by 38 students (by using two tests with 08 open questions each). The analisys of the results with Pearson correlation shows strong association (0,81) between grades given by SAICOweb and the human grades (corrected by professors). Another important result is the satisfaction of professor and students when evaluating Software SAICOweb, evaluation method, pedagogical experience and anonymity. Specific itens were identified and suggested as future updates, when more experimental tests should be designed to validade other aspects of the System.