Evaluation of mathematical questioning strategies using data collected through weak supervision

A large body of research demonstrates how teachers' questioning strategies\ncan improve student learning outcomes. However, developing new scenarios is\nchallenging because of the lack of training data for a specific scenario and\nthe costs associated with labeling. This paper presents a high-fidelity,\nAI-based classroom simulator to help teachers rehearse research-based\nmathematical questioning skills. Using a human-in-the-loop approach, we\ncollected a high-quality training dataset for a mathematical questioning\nscenario. Using recent advances in uncertainty quantification, we evaluated our\nconversational agent for usability and analyzed the practicality of\nincorporating a human-in-the-loop approach for data collection and system\nevaluation for a mathematical questioning scenario.\n