A Platform-Agnostic Model and Analysis of Learner Engagement within Peer Supported Digital Environments: FutureLearn MOOCs and PeerWise

Abstract

Digital technologies have accelerated a conceptual shift in education from traditional face-to-face instruction towards an increasingly asynchronous, online, learner-centred paradigm. Under this paradigm, learners interact both with peers and content matter, leaving traces that can be used to characterise their learning engagement. This is the focus of a growing interest in learning analytics, particularly with data mining algorithms, of which clustering are an important class. These algorithms are however usually applied to datasets from a single platform, leading to platform-specific findings.
This thesis presents a new model of learner engagement within peer-supported digital environments that describes interactions independently of their platform, and can help make meaningful comparisons across contexts. The model was validated by ap- plying a machine-learning approach to datasets from courses in face-to-face instruction and online. Data processed were from a total of 271,851 learners from nineteen courses from the University of Southampton between 2014-2019 on topics on archaeology, lan- guage teaching and human-computer interaction. Seventeen of these were massive open online courses (MOOCs), and the remaining two were in a face-to-face setting that included the use of PeerWise as a peer-supported digital environment.
Feature engineering was performed on timestamped digital traces of activity using this new model, producing sixteen feature files with up to 78 features per learner, which were subjected to the clustering algorithms Expectation Maximization, Simple k-Means and X-Means with k values varying from two to ten. Highly-interpretable clusters were identified by X-Means on dialogic features from datasets from both platforms, allowing for a meaningful comparison of learner engagement across environments. In particular, engagement in both platforms was found to fall in four main activity classes ranging from asocial to fully active social learners; although nuanced behaviours were also evi- denced. Learning design was found to affect the composition of these clusters, and when free of behavioural constraints, learners in the face-to-face environment evidenced the same types of behaviours as those online.

More information

Identifiers

ORCID for Adriana Wilde: orcid.org/0000-0002-1684-1539

ORCID for David Millard: orcid.org/0000-0002-7512-2710

Catalogue record

Export record