Human-in-the-Loop LLM Workflow for Trustworthy Qualitative Data Analysis
A workflow that couples a large language model with human interpretive judgement to perform qualitative data analysis - coding text into sub-themes and themes - while keeping the process rigorous and accountable. It uses chain-of-thought prompting, LLM self-critique surfaced for human review, and a full audit trail of decisions and revisions, so the model proposes but the human decides, producing a traceable, shareable codebook.
2501.00775
Qualitative data analysis (QDA) turns unstructured text into codes, sub-themes and themes through iterative, interpretive human judgement - rigorous but slow. This paper presents MindCoder, a human-i…