Abstract:

Massive open online courses (MOOCs) are a widely available, low-cost platform to provide collaborative workspaces of online educational delivery that supply students with online courses and assessments to strengthen individual participation and joint knowledge development. In the last few years, MOOCs have proliferated mainly due to information and communication technological advances and the learners' widespread use of such technologies, which have removed the geographical barriers for participants. Thus, MOOCs currently have a significant number of registered learners. However, a significant percentage of learners leave the courses and drop out of the platform.

MOOC platforms in general collect, register, and monitor data related to courses, classes, students, resource usage, and interactions. For example, these platforms collect stats about the video watching (e.g., time watching the video or number of times the video was paused), the assignments (e.g., homework scores and the provided solutions), and the class forum discussions (e.g., number of students participating or number of posts). This gathered data represents a valuable source of information that should be exploited by the agents involved in the educational process. Thus, educational data mining (EDM) is an emerging research field focused on applying computational intelligence tools and techniques from data mining, machine learning (ML), and statistics to data collected from educational settings, with the aim of better understanding students and the settings of the learning process.

This article surveys recent literature on the application of ML to exploit MOOC gathered data to extract meaningful patterns and discover useful knowledge about study behavior. We collected journal articles of the last five years from the Scopus, ACM, IEEE XPlore and Web of Science databases with more than 300 entries. Specific exclusion and inclusion criteria were applied following PRISMA guidelines to select the most relevant studies. The final 40 evaluated research studies apply and or combine ML techniques and methods, such as natural language processing (NLP), supported vector machines (SVM), Long short-term memory (LSTM), k-Nearest Neighbor (k-NN), or deep neural networks (DNN), among others, to address and support students’ behavior understanding while using MOOCs. These are classified across 4 categories: Process (cognitive level of engagement with study activities), Strategies (cognitive level of control over study activities), Habits (consistency and actualization of study activities) and Tactics (learning tools used by students). Overall, this systematic literature review illustrates the main trends on this research topic.

Keywords:

MOOC, Machine Learning, Study Behaviour.