Exploring Effective Strategies for Building a Customised GPT Agent for Coding Classroom Dialogues

A study evaluated a custom GPT agent for classroom dialogue analysis, aiming to automate manual coding. The agent, utilizing GPT-4's MyGPT, showed promise as a coding assistant, with performance notably improving with more training data. Contextual dialogue samples outperformed single-turn ones, and focused examples benefited structurally consistent codes. The study emphasized the importance of data structure, diversity, and dialogic richness for model performance. It proposed strategies for creating an effective MyGPT model for coding dialogues. Four studies explored AI in qualitative research and education, focusing on dialogic practices, GPT-3 for coding, and cognitive load management. Two studies discussed large language models, including QualiGPT's potential and conformity issues.