Monitoring Programming Styles in Massive Open Online Courses Using Source Embedding

Abstract

In recent years there has been an exponential growth of distance learning, provided by both public and private institutions. As a matter of fact, the number of students enrolled in courses delivered through the Network, has dramatically grown, also due to the COVID-19 pandemic, which has forced millions of people not to move. Consequently, more and more courses delivered in a remote modality have been attended by a huge number of people, producing an increasing number of Massive Open Online Courses (MOOC)s. These kind of courses are imposing new challenges for teachers, especially for monitoring and assessing the community learning processes. On the one hand, the learning assessment cannot be carried out based solely on closed-ended tests, while, on the other hand, teachers cannot evaluate thousands of open-answer assignments: they should have at their disposition a set of tools helping them monitor the community learning progress. This paper investigates the possibility of using some of the Source Code Embedding techniques, to give teachers useful information about their learners' programming styles in Massive Open Online Courses. We propose a method to visualize each student's program, included the teacher's one, as a point in a 2-D space, using the doc2vec embeddings technique. Thanks to this representation, teachers can identify in the 2-D space groups of students having similar programming styles and reason on them to start a suitable didactic feedback. Moreover, teachers can reason on the relationship between each point compared to their own point as well, considered as the truth programming style. A first experimentation using Python as the programming language is performed with encouraging results.