Automating Thematic Analysis Using Large Language Models

Abstract

Thematic analysis is a foundational qualitative research method widely used across social sciences, healthcare, education, and policy studies to identify, analyze, and interpret patterns within textual data. Despite its methodological importance, traditional thematic analysis remains time-intensive, subjective, and difficult to scale, particularly in the era of big data. Recent advances in Large Language Models (LLMs) present a compelling opportunity to automate and augment this process. This paper investigates the feasibility, effectiveness, and limitations of automating thematic analysis using LLMs. We propose an end-to-end framework that leverages transformer-based language models for theme extraction, clustering, and semantic coherence evaluation. Through extensive experimentation on benchmark qualitative datasets, we demonstrate that LLM-driven thematic analysis achieves high alignment with human-coded themes while significantly reducing analysis time. Our results suggest that LLMs can serve as reliable collaborative agents for qualitative research rather than mere automation tools, opening new pathways for scalable, reproducible, and explainable qualitative analysis