DeTAILS: Deep Thematic Analysis with Iterative LLM Support
Fuente:
arXiv
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911224162156544 |
|---|---|
| author | Sharma, Ansh Cochrane, Karen Wallace, James R. |
| author_facet | Sharma, Ansh Cochrane, Karen Wallace, James R. |
| contents | Thematic analysis is widely used in qualitative research but can be difficult to scale because of its iterative, interpretive demands. We introduce DeTAILS, a toolkit that integrates large language model (LLM) assistance into a workflow inspired by Braun and Clarke's thematic analysis framework. DeTAILS supports researchers in generating and refining codes, reviewing clusters, and synthesizing themes through interactive feedback loops designed to preserve analytic agency. We evaluated the system with 18 qualitative researchers analyzing Reddit data. Quantitative results showed strong alignment between LLM-supported outputs and participants' refinements, alongside reduced workload and high perceived usefulness. Qualitatively, participants reported that DeTAILS accelerated analysis, prompted reflexive engagement with AI outputs, and fostered trust through transparency and control. We contribute: (1) an interactive human-LLM workflow for large-scale qualitative analysis, (2) empirical evidence of its feasibility and researcher experience, and (3) design implications for trustworthy AI-assisted qualitative research. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_17575 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | DeTAILS: Deep Thematic Analysis with Iterative LLM Support Sharma, Ansh Cochrane, Karen Wallace, James R. Human-Computer Interaction Thematic analysis is widely used in qualitative research but can be difficult to scale because of its iterative, interpretive demands. We introduce DeTAILS, a toolkit that integrates large language model (LLM) assistance into a workflow inspired by Braun and Clarke's thematic analysis framework. DeTAILS supports researchers in generating and refining codes, reviewing clusters, and synthesizing themes through interactive feedback loops designed to preserve analytic agency. We evaluated the system with 18 qualitative researchers analyzing Reddit data. Quantitative results showed strong alignment between LLM-supported outputs and participants' refinements, alongside reduced workload and high perceived usefulness. Qualitatively, participants reported that DeTAILS accelerated analysis, prompted reflexive engagement with AI outputs, and fostered trust through transparency and control. We contribute: (1) an interactive human-LLM workflow for large-scale qualitative analysis, (2) empirical evidence of its feasibility and researcher experience, and (3) design implications for trustworthy AI-assisted qualitative research. |
| title | DeTAILS: Deep Thematic Analysis with Iterative LLM Support |
| topic | Human-Computer Interaction |
| url | https://arxiv.org/abs/2510.17575 |