Abstract

Interest in studying Massive Online Open Courses (MOOC) learners' sessions has grown as a result of the retention and completion issues that these courses present. Applying process mining to study this phenomenon is difficult due to the freedom of navigation that these courses give their students. The goal of this research is to provide a domain-driven top-down method that enables educators who are unfamiliar with data and process analytics to search for a set of preset high-level concepts in their own MOOC data, hence simplifying the use of typical process mining techniques. This is accomplished by defining a three-stage process that generates a low-level event log from a minimum data model and then abstracts it to a high-level event log with seven possible learning dynamics that a student may perform in a session. By examining the actions of students who successfully completed a Coursera introductory programming course, the framework was tested. As a consequence, patterns in the repetition of content and assessments were described; it was discovered that students' willingness to evaluate themselves increases as they advance through the course; and four distinct session types were characterized via clustering. This study shows the potential of employing event abstraction strategies to gain relevant insights from educational data.