Knowledge of prerequisite dependencies is crucial to several aspects of learning, from the organization of learning content to the selection of personalized remediation or enrichment for each learner. As the amount of content is scaled up, however, it becomes increasingly difficult to manually specify all of the prerequisites among the different content parts, necessitating automation. Since existing approaches to automatically inferring prerequisite dependencies rely on analysis of content (e.g., topic modeling of text) or performance (e.g., quiz results tied to content) data, they are not feasible in cases where courses have no assessments or only short content pieces (e.g., short video segments). In this paper, we propose an algorithm that extracts prerequisite information using learner behavioral data instead of content and performance data, and apply it to an online short course. By modeling learner interaction with course content through a recurrent neural network-based architecture, our algorithm characterizes the prerequisite structure as latent variables, and estimates them from learner behavior. Through evaluation on a dataset of roughly 12,000 learners in a course we hosted on our platform, we show that our algorithm excels at both predicting behavior and revealing fine-granular insights into prerequisite dependencies between content segments, with validation provided by a course administrator. Our approach of content analytics using large-scale behavioral data complements existing approaches that focus on course content and/or performance data. [For the full proceedings, see ED593090.]