TOPOLOGY OF DYNAMIC NETWORKS OF KNOWLEDGE OF MICROLEARNING CONTENT: TOPIC MODELING APPROACHES

One of the issues faced when generating personalized learning paths (PLPs) is the organization and tagging of microlearning content. We expand upon network science approaches from previous work in Dynamic Network of Knowledge (DNoK) design by incorporating topic modeling methods to automate keyword extraction. Specifically, we compare the BERT language model with fast keyword extraction methods to assign keywords in common using the original DNoK framework. For instructors, institutions’ content curators, and designers of topic-specific learning management systems, we formally define a minimum viable network statistic from which DNoK growth should occur. We continue our comparison of topological growth between networks produced from topic modeling and manually curated keyword methods from previous work. Finally, we examine the growth behavior of the DNoK and propose bounds to support the generation of synthetic models for exploring such networks at scale. DNoK design differs from bibliometric and other keyword co-occurrence networks in that those co-occurrences are embedded in edge weights instead of node degree. This framework enables discovery of content and grouping of learning concepts through traditional community detection methods. Ultimately, we propose that a network science-based approach will facilitate personalized, adaptive learning methods that enable instructors and learning engineers to integrate best-practices from learning and cognitive science within this space.