LLM-assisted thematic analysis in empirical legal studies

Research applicability

Offers a concrete qualitative-analysis research use case for empirical legal scholars, with expert feedback retained rather than replaced.

Getting started

Read https://arxiv.org/abs/2310.18729 (inspected v1) for the study framing; inspect full methods and data permissions before adaptation. Suggested adoption steps are editorial guidance, not a tested workflow.

Maintenance

Static research publication; current software maintenance not assessed.

License

License not audited; consult source terms before reuse. This is not an exclusion criterion.

Workflow & human oversight

Defined researcher-directed coding/theme-analysis workflow; human interpretation and checking remain necessary. No autonomous research loop established.

Workflow level is not an evidence score; higher is not better evidence.

Approach

A legal expert and LLM collaborate across initial codes, theme search and classification; the authors report code improvement with expert feedback and correspondence with expert-derived themes.

Limitations

Results concern one theft-related court-opinion dataset and GPT-4; transfer to other jurisdictions, offences or models is not established. Primary-source abstract and metadata review only; full-text methods, implementation and data were not inspected. No model or experiment was run and no outcome independently reproduced.

Reproducibility

Primary-source abstract and metadata review only; full-text methods, implementation and data were not inspected. No model or experiment was run and no outcome independently reproduced.

Adoption requirements

For adaptation, access to a suitable LLM, lawfully usable court-text data and a legal researcher able to check codes, themes and contextual meaning; prompts and implementation were not reproduced.