Uncovering latent features in massive open online courses

Thesis

Massive Open Online Courses (MOOCs) bring together a global crowd of thousands of learners for several weeks or months. We use social network analysis and community detection to uncover the latent features of online discussions in MOOCs. We begin by using data from two successive instances of a popular business strategy MOOC to filter observed communication patterns and arrive at the "significant" interaction networks between learners. We then use complex network analysis to explore the vulnerability and information diffusion potential of the discussion forums. While network analysis offers a vibrant post-hoc analytical framework, it fails to answer a fundamental question: can we devise a model to represent the generation of the dataset at hand?

Moving from the structural properties of global-scale discussion to the discussion content itself, we employ existing educational theories to qualitatively content-analyse over 6,500 forum posts from a particular MOOC. We then use a generative model - Bayesian Non-negative Matrix Factorization (BNMF) - to extract communities of learners based on the nature of their online forum posts. We observe that the inferred communities are differentiated by the nature and topic of dialogue, as well as their composite students' demographic and course performance indicators. While qualitative analysis confirms these detected communities, additional quantitative sensitivity analysis shows that they are not crisply defined, illuminating key challenges of applying Machine Learning techniques to model noisy and incomplete learner data.

We conclude by discussing the key insights of this work for online education, namely, that different discussion topics and pedagogical practices promote varying levels of peer-to-peer engagement. Additional qualitative investigations reveal that many learners feel a sense of "content-overload" when deciding to participate in online discussions, often leading to their disengagement. These insights call for an interdisciplinary effort to help create relevant and personalized learning experiences in massive scale online settings.

Authors: Gillani, N

• Publication date: 2014
• Type of award: MSc by Research
• Level of award: Masters
• Awarding institution: University of Oxford
• Language: English
• Keywords: network analysis; Machine learning; online learning; education
• Subjects: Education; Pattern recognition (statistics); e-Learning; Probability; Computer science (mathematics)