Further explorations on the use of large language models for thematic analysis: open-ended prompts, better terminologies and thematic maps

In this manuscript I build upon an initial body of research developing procedures for leveraging large language models (LLMs) in qualitative data analysis, by carrying out thematic analysis (TA) with LLMs. TA is used to identify patterns by means of initial labelling of qualitative data followed by the organisation of the labels/codes by themes.

First, I propose a new set of LLM prompts for initial coding and generation of themes. These new prompts are different from the typical prompts deployed for such analysis in that they are entirely open-ended and rely on TA language. Second, I investigate the process of removing duplicate initial codes through a comparative analysis of the codes of each interview against a cumulative codebook. Third, I explore the construction of thematic maps from the themes elicited by the LLM. Fourth, I evaluate the themes produced by the LLM against the themes produced manually by humans. For conducting this research, I employed a commercial LLM via an application program interface (API). Two datasets of open access semi-structured interviews were analysed to demonstrate the methodological possibilities of this approach. I conclude with practical reflections on performing TA with LLM, enhancing our knowledge of the field.