Detecting the Depth and Progression of Learning in Massive Open Online Courses by Mining Discussion Data


In massive open online courses (MOOCs), learners can interact with each other using discussion boards. Automatically inferring the states or needs of learners from their posts is of interest to instructors, who are faced with a high attrition in MOOCs. Machine learning has previously been successfully used to identify states such as confusion or posting questions, but no solution has yet been provided so that instructors can track the progress of the learners using a validated framework from education research. In this paper, we develop a model to automatically label a post based on the first phase of the interaction analysis model (IAM). This allows instructors to automatically identify whether students are stating opinions, clarifying details, or engaging in activities such as providing examples to peers. Our model is tested on a Coursera MOOC devoted to Chemistry, for which we are able to correctly categorize the IAM status in 4 out of 5 posts. Our approach thus provides instructors with an intelligent system that generates actionable learning assessment data and can cope with large enrollment. Using the system, instructors can quickly identify and remedy learning issues, thus supporting learners in attaining their intended outcomes.

Publication Title

Technology, Knowledge and Learning