Human-LLM Collaborative Coding Pipeline for K-12 Educator AI Interaction Analysis

There's a new preprint on arXiv (2607.28889) that introduces a team-based approach combining humans and large language models (LLMs) to analyze a massive dataset of 45,000 messages between K-12 teachers and an AI system. The researchers utilized various coding methods to build a structured codebook, where LLMs suggest labels while human researchers decide on definitions and interpretations. The effectiveness of this tool was tested with thorough coding by three trained experts. This study looks at how LLMs can aid qualitative research and proposes a model that balances automation with human oversight. You can find this preprint categorized as a cross-type announcement on arXiv.