LLM-TA: LLM-Enhanced Thematic Analysis Pipeline

Code for the paper "LLM-TA: An LLM-Enhanced Thematic Analysis Pipeline for Transcripts from Parents of Children with Congenital Heart Disease" (Accepted by GenAI4Health Workshop at AAAI 2025).

Getting Started

git clone https://github.com/jiaweixu98/LLM-TA.git

To reuse the code, you need to provide your LLM API key. Refer to the LangChain official documentation for setup instructions.

Note: The transcripts are not included in this repository. You can use your own data and modify the prompt to fit your context.

Notebooks

• 0_shot.ipynb: Contains our proposed method and evaluation code in a direct input/output setting.
• 1_shot.ipynb: Contains our proposed method and evaluation code with one exemplar provided.
• reflexion.ipynb: Contains our proposed method and evaluation code using the Reflexion strategy.

Baselines

The baselines/ directory includes two baselines from:

De Paoli, S. (2024). Performing an inductive thematic analysis of semi-structured interviews with a large language model: An exploration and provocation on the limits of the approach. Social Science Computer Review, 42(4), 997–1019.

Citation

If you find this paper useful, please cite our paper:

@misc{raza2025llmtallmenhancedthematicanalysis,
      title={LLM-TA: An LLM-Enhanced Thematic Analysis Pipeline for Transcripts from Parents of Children with Congenital Heart Disease}, 
      author={Muhammad Zain Raza and Jiawei Xu and Terence Lim and Lily Boddy and Carlos M. Mery and Andrew Well and Ying Ding},
      year={2025},
      eprint={2502.01620},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2502.01620},
      note={Accepted at GenAI for Health Workshop, AAAI 2025, Philadelphia}
}