An Adaptive Content Skipping Methodology based on User Behavioral Modeling

Adaptive Educational Systems (AES) have demonstrated the potential of improving learning efficacy by individualizing course delivery to particular user needs. Whereas the algorithms driving today's AES are primarily based on user responses to quiz questions, these systems are now capable of capturing fine-granular behavioral data on users, such as keystroke measurements on lecture videos and social interactions on discussion forums. In this paper, we develop a methodology that leverages behavioral data for the task of content skipping in an AES, i.e., detecting content segments that are unnecessary for a user and passing over them automatically. Our methodology contains three modules: (1) a Behavioral Data Processor, which converts user behaviors and course content into algorithm features including course topics, (2) a User State Tracer, which maintains an estimate of user knowledge state and interest on a per-topic basis, and (3) a Content Skipping Trigger, which determines the segments to be removed from the course for this user. In evaluating our approach on two real-world datasets collected from courses hosted on our existing platform, we find 80-90% accuracy in terms of identify segments that users would themselves eventually skip. In doing so, we also perform some exploratory analysis to show how the prediction results can help instructors to improve the course design quality.