Feedback dos alunos sobre a qualidade de código-fonte para apoiar o aprendizado de introdução à programação.

Abstract

Introductory programming subjects typically involve a large number of assignments, which makes it costly to provide feedback for each student throughout the term manually. In this way, several approaches are proposed to provide automatic feedback on the solutions submitted by the students. The focus of these researches is on functional feedback. That is if the program is correct according to predefined tests by teachers. However, there is also a need to analyze the code quality produced by the students. Although there are research efforts in this direction, the proposed approaches focus on the automated analysis of syntactic aspects, such as the criteria proposed in the PEP 8, but which can lead to general feedback. In this study, we investigated whether, by including students as reviewers, we could provide personalized feedback on the code quality. Thus, in this dissertation, the central question is whether students can qualitatively evaluate the program of your pairs. For this, we carried out a study to verify if the students’ hints are useful and similar to those of the teachers of the subject. We found that most students can identify code quality issues related to the teachers at a significant level, even if not ideal. We have also seen that students can give meaningful hints at a considerable level, and are particularly able at finding and giving hints on issues related to programs complexity. We replicated this study in another context and obtained similar results to corroborate what we observed. After we found that students can give hints, we conducted a peer code review experiment. We found that students can be specific and propose alternatives to issues identified. However, most advanced students of the subject can give more explanatory hints. Finally, we applied quality metrics in the codes before and after the feedback and found that there was an improvement, but we did not get enough data to affirm that it was significant statistically. We verified that the student’s knowledge influences in its result. However, we observed that experience in providing feedback and the motivation are also performance-impacting factors. This study may lead to further research on how to approach the code quality in collaborative learning and more advanced programming courses.