Domain-Driven Event Abstraction Framework for Learning Dynamics in MOOCs Sessions

Abstract

In conjunction with the rapid expansion of Massive Open Online Courses (MOOCs), academic interest has grown in the analysis of MOOC student study sessions. Education researchers have increasingly regarded process mining as a promising tool with which to answer simple questions, including the order in which resources are completed. However, its application to more complex questions about learning dynamics remains a challenge. For example, do MOOC students genuinely study from a resource or merely skim content to understand what will come next? One common practice is to use the resources directly as activities, resulting in spaghetti process models that subsequently undergo filtering. However, this leads to over-simplified and difficult-to-interpret conclusions. Consequently, an event abstraction becomes necessary, whereby low-level events are combined with high-level activities. A wide range of event abstraction techniques has been presented in process mining literature, primarily in relation to data-driven bottom-up strategies, where patterns are discovered from the data and later mapped to education concepts. Accordingly, this paper proposes a domain-driven top-down framework that allows educators who are less familiar with data and process analytics to more easily search for a set of predefined high-level concepts from their own MOOC data. The framework outlined herein has been successfully tested in a Coursera MOOC, with the objective of understanding the in-session behavioral dynamics of learners who successfully complete their respective courses.