Recomendação de exercícios para alunos de programação em um ambiente de correção automática de códigos

Abstract

Many students in Introductory Programming (CS1) courses have difficulty learning to program. Therefore, programming teachers have used online judges to propose exercises, run marathons and programming championships, in order to try to improve the students' learning experience. However, in cases of online judges who have many registered exercises in their database, the student may choose an exercise that is not suitable for their level of knowledge. In this sense, this work proposes an exercise recommendation system, which filters exercises by level of difficulty, in an online judge called CodeBench. These exercises are classified manually by subjects by the teacher, and the method proposed in this research suggests exercises ordered by level of difficulty. For this, the collaborative filtering recommendation approach is used to map the difficulties experienced by students when solving programming exercises in the CodeBench integrated development environment. After that, the prediction of the difficulty of the exercises that the student has not yet solved is made to then suggest exercises with increasing degrees of difficulty. This recommendation method was applied to a database of 645 undergraduate students, distributed in 14 CS1 classes, taught in 2018, at a public university. The course is divided into 7 modules, each module has two lists of exercises and a test. In each module, the original order of exercise resolution was compared with the order of exercise resolution proposed by the method. The results show that in 6 modules of the discipline the method proposed here suggests an order of resolution adapted for each student, with increasing level of difficulty.