Improving Course Design Through LLM-Assisted Educational Process Mining

Abstract

Learning analytics increasingly draws on digital traces from learning management systems to understand how students engage with online courses. Educational process mining (EPM) derives process models from event logs, offering a process-oriented view of behavior, but models built from large cohorts and rich activity data often become complex and difficult to interpret. This chapter presents an analytical workflow that integrates EPM with large language models (LLMs): event logs are used to discover a process model, the model is transformed into a structured textual representation, and an LLM is then used to summarize structures and surface salient patterns. The workflow is illustrated in an undergraduate course, using interaction data from a cohort of low-performing students in the institutional virtual campus. GPT-5 is employed to interpret the discovered model, revealing hub-and-spoke navigation centered on the course page, frictions in quiz completion, and non-linear progress in peer-assessment activities. Based on these findings, we derive concrete course-design recommendations aimed at streamlining navigation, clarifying assessment closure, and improving guidance in complex tasks, and we discuss how LLM-assisted EPM can help address scalability and interpretability challenges in adaptive learning contexts.