Clustering of learners' behaviour in the Understanding Language MOOC

Discussions on success of learners in Massive Open Online Courses (MOOCs) have long moved beyond a simplistic view of completion rates alone and are increasingly taking into account the motivations of learners in pursuing a MOOC. There have been endeavours in characterising the diversity of learners in a handful of "archetypes" (Walker, 2018) which aid understanding both the motivations and needs of participants falling in these categories. In principle, this is only possible by collating and analysing self-reported data on learners’ motivation, together with their actual behaviour in the platform. However, in practice, such self-reported data is rare in comparison with the wealth of data available on learners’ interaction with and within the platform. "Sub-populations" (Kizilcec et al., 2013; Ferguson & Clow, 2015) can still be identified only by observing the behaviour in the platform, and arguably represent a very similar classification of learners to that in the archetype analysis. This talk presents our contribution to this debate. We have studied learners’ engagement in the 5-week FutureLearn course "Understanding Language" on its first six offerings (from 2014-2017), facilitated by the British Council in Collaboration with the University of Southampton. Using three clustering algorithms on the related datasets with only step-activity, enrolments and comments (including number of likes), we have identified six clusters: Samplers, Strong Starters, Unsocial Learners, Popular, Fully Engaged and Atypical Learners. Samplers take the largest part of learners in all runs of the course, as expected according to the Funnel of Participation model (Clow, 2013).