Leveraging LLMs on Teacher Voices to Drive Retention Policies

• In Event: AI, Machine Learning, and Data Science in Modeling and Prediction: New Strategies and Applications

Abstract

This study explores how large language models (LLMs) can be used to transform open-ended responses from a statewide teacher survey into reliable, policy-relevant indicators of job demands and resources. Using the Job Demands–Resources (JD-R) framework, we evaluated multiple LLM prompting strategies for classifying qualitative responses by theme and sentiment, comparing model outputs to expert human coding. The optimal prompt will be used to code over 8,000 comments from the 2023–24 Teacher Working Conditions Survey (TWCS). We will then conduct bias and sensitivity testing to assess model performance across school poverty levels and geographic locales. Findings will offer practical guidance on applying LLMs to elevate educator voice and inform data-driven strategies for teacher retention and support.

Authors

• Angela D. Starrett, University of South Carolina

• Svetlana Dmitrieva, University of South Carolina

• Brian Cartiff, University of South Carolina