Edu-Theater Simulates Learner Behavior with Data Efficiency

Collecting large-scale learner-task interaction data is essential for intelligent educational systems but is often expensive and limited by privacy concerns and learner engagement. Learner simulators are valuable for generating scalable learner behavior without continuous real-learner involvement. However, current methods are typically individual-centric, requiring extensive data and computation for each learner, making them fragile in new scenarios. A new approach, Edu-Theater, introduces a cohort-aware roll-call simulation paradigm. This system first establishes proficiency priors at the cohort level and then refines individual learner states using a small number of targeted diagnostic queries. Edu-Theater is an LLM-powered agent system that performs this cohort-aware simulation through a teacher agent and retrospective roll-call probing of learner logs. This framework enables scalable future behavior simulation without needing dense historical data for each learner. Experiments show that Edu-Theater achieves higher simulation accuracy with significantly fewer LLM calls and produces synthetic data that enhances applications like adaptive testing.