Modelos e técnicas para melhorar a qualidade da avaliação automática para atividades escritas em língua portuguesa brasileira

Abstract

Against the workload generated by the evaluation of activities written in online learning, different systems, like Coursera and edX, have adapted their approaches to the evaluation of the appropriate activities. However, using differents approaches, both systems do use of the most popular methods of evaluating such activities, peer review, and automatic evaluation. However, each technique presents its advantages and disadvantages, especially under the student learning perspective, since in addition to providing an increase in learning, peer evaluation stands out for having quality in the assessment similar to that of a specialist. But, automatic essay evaluation is a method that provides fast feedback, and how larger the training base is, the technique has better efficacy, in addition to being used with the online learning system in different aspects, such as monitoring the existing limitations in students, among other words, what are the doubts that a student has, which are extracted from patterns derived from the evaluation of subjective activities. Thus, the present work proposes a model to evaluate efficiently subjective activities, reducing the workload generated by the evaluation of subjective activities by the teacher. The automatic essay score system was calibrated with approximately 5407 evaluations and essays, where it underwent an experiment containing 60 essays which indicated that the evaluations carried out by the approach are similar to the expert’s evaluations with 95% level of trust. In addition to showing a higher accuracy in relation to the evaluation provided by the approach.