Analysis code and response-level dataset for: An Interpretable Machine Learning Framework for Automated Screening of Patient-Facing LLM Consultations (Medical Aesthetic Consultation)
This deposit contains the complete analysis pipeline (Python) and the response-level dataset (600 LLM-generated responses with blinded expert ratings, readability metrics, and experimental design variables) supporting the manuscript "An Interpretable Machine Learning Framework for Automated Screening of Patient-Facing LLM Consultations: A Comparative Evaluation in Medical Aesthetic Consultation" (Scientific Reports, Submission ID 170dac52-b32e-4486-aca1-739ce0814650). The dataset comprises 600 responses in a fully crossed design (25 seed questions x 6 prompt strategies x 2 query modes x 2 LLMs: DeepSeek-V3 and Kimi). The code reproduces all t
Resource information
- Format
- zip
- Original repository
- 23559222
- DOI
- 10.5281/zenodo.23559222
- Credit basis
- Creators in Zenodo record metadata
- Access
- Open files available from the original Zenodo deposit
- Last checked
- Nov 5, 2026
- License
- cc-by-4.0
- Last record update
- Nov 5, 2026
About this resource
This deposit contains the complete analysis pipeline (Python) and the response-level dataset (600 LLM-generated responses with blinded expert ratings, readability metrics, and experimental design variables) supporting the manuscript "An Interpretable Machine Learning Framework for Automated Screening of Patient-Facing LLM Consultations: A Comparative Evaluation in Medical Aesthetic Consultation" (Scientific Reports, Submission ID 170dac52-b32e-4486-aca1-739ce0814650). The dataset comprises 600 responses in a fully crossed design (25 seed questions x 6 prompt strategies x 2 query modes x 2 LLMs: DeepSeek-V3 and Kimi). The code reproduces all tables and figures, including the question-level grouped cross-validation (primary analysis of the revised manuscript) and threshold sensitivity analyses. See README.md inside the archive for details. Version 3 adds the round-3 revision analyses: paired between-model tests (Wilcoxon signed-rank, Friedman), applicability-adjusted composite sensitivity analyses, and triage-aligned evaluation (precision–recall and review-burden operating points).