ChikungunyaQA: A High-Fidelity, Multi-Persona Clinical Question-Answering Dataset Derived from Brazilian Ministry of Health Guidelines

ChikungunyaQA: A Clinical Dataset of Questions and Answers on Chikungunya ChikungunyaQA is a multi-persona question-and-answer (QA) dataset in Brazilian Portuguese (PT-BR) derived from official clinical guidelines published by the Brazilian Ministry of Health and complementary regulatory documents. Dataset Overview - Total QA Pairs:** 1,262- Language:** Brazilian Portuguese (PT-BR)- Personas:** - Physician (Médica):** 556 pairs (44.1%) — formal medical terminology, dosage guidelines, and clinical workflows. - Patient (Paciente):** 581 pairs (46.0%) — layperson language, accessible terms, and general symptom identification. - Caregiver (Cuidador):** 125 pairs (9.9%) — practical home care, monitoring warning signs, and patient support.- Validation:** Dual-evaluation framework combining an LLM judge (G-Eval with Claude Sonnet 4.5, mean score: 97.1/100) and human expert evaluation by healthcare professionals (a physician and a physical therapist).- License:** Creative Commons Attribution 4.0 International (CC BY 4.0) Dataset Formats & Files 1. `ChikungunyaQA.jsonl` — Full Research Format (JSONL): Preserves complete structured metadata (source document, page numbers, disease stage, semantic tags, dimensional G-Eval scores, and qualitative judge reasoning). Ideal for auditing, benchmarking, and semantic retrieval in RAG.2. `ChikungunyaQA.csv` — Full Research Format (CSV): Preserves complete structured metadata in tabular CSV format for spreadsheet analysis and benchmarking.3. `ChikungunyaQA_alpaca.jsonl` — **Alpaca Format for SFT Training (JSONL):** Preprocessed instruction-style version (`instruction`, `input`, `output`, `persona`), ready for direct fine-tuning of language models via techniques such as LoRA or QLoRA. FAIR Principles Compliance This dataset complies with the FAIR principles (*Findability, Accessibility, Interoperability, and Reusability*):- Findable: Assigned permanent DOI `10.5281/zenodo.20444766` and structured metadata.- Accessible: Publicly accessible on Zenodo under open CC BY 4.0 license.- Interoperable: Distributed in standard open formats (JSONL, CSV) compatible with Hugging Face Datasets, Unsloth, LLaMA-Factory, and pandas.- Reusable: Enriched with comprehensive clinical provenance, metadata, and clear license terms for model fine-tuning (SFT) and Retrieval-Augmented Generation (RAG) benchmarking. Citation If you use ChikungunyaQA in your research, please cite: ```bibtex@article{chikungunyaqa2026, title={ChikungunyaQA: A Clinical Dataset of Questions and Answers on Chikungunya}, author={dos Santos, Denival Araujo and Moreira, Rayele and da Costa, Ana Cristina Vieira and de Barros, Gabriel Martins and Santos, Bruno Sampaio and Teles, Ariel Soares}, journal={Data}, year={2026}, doi={10.5281/zenodo.21908616}}```