Loading...

 

What is it about?

Thematic analysis is a key method in qualitative research, where researchers review multiple interview transcripts to identify common themes. This process, however, can be slow and require significant effort. GPT models can help streamline this work, but they face challenges such as handling limited input sizes (known as the "context window") and sometimes overlooking texts situated in the middle of lengthy inputs (known as the "lost in the middle" phenomenon). To address this, we leverage advanced prompting techniques to develop tools which enable theme extraction across multiple lengthy transcripts. When tested on 11 studies using four criteria—clarity, alignment with researcher-identified themes, explanation quality, and relevance of quotes—the approach achieved an average score of 3.05 out of 4. This shows it can effectively identify themes while reducing the workload for researchers. While the tool is helpful, we recommend pairing its outputs with domain-specific expertise to ensure high-quality insights. Researchers can try our open-source Python package, AutoThemeGenerator, available at https://pypi.org/project/AutoThemeGenerator/

Featured Image

Why is it important?

Performing thematic analysis is often a labor-intensive process, requiring researchers to analyze multiple lengthy transcripts to identify key themes. While Generative AI tools can assist qualitative researchers, concerns remain about their tendency to produce inaccurate or unreliable results, often referred to as hallucinations. Additionally, these tools face technical limitations, such as restricted input capacity. To address these issues, we developed an open-source Python package designed to overcome these limitations. Through rigorous evaluation, we provide qualitative researchers with a user-friendly and reliable tool that has been carefully assessed for its safety and effectiveness.

Perspectives

RASorry, your browser does not support inline SVG.

By sharing our open-source Python package, AutoThemeGenerator (https://pypi.org/project/AutoThemeGenerator/), which has been rigorously evaluated, we aim to support qualitative researchers in the public health field. Our goal is to help them leverage Generative AI tools to enhance research productivity and streamline thematic analysis.

Ruopeng An
New York University

Read the Original

This page is a summary of: GPT Models Can Perform Thematic Analysis in Public Health Studies, Akin to Qualitative Researchers, Journal of Social Computing, December 2024, Tsinghua University Press,
DOI: 10.23919/jsc.2024.0024.
You can read the full text:

Read

Resources

Contributors

The following have contributed to this page