Datasets:

Xueren
/

EMS-Knowledge

dataset_name: EMS-Knowledge
pretty_name: EMS Knowledge Base (Flattened Sections)
language:
  - en
license: mit
tags:
  - emergency-medical-services
  - ems
  - medical
  - knowledge-base
  - retrieval
  - question-answering
size_categories:
  - 10K<n<100K
task_categories:
  - question-answering
  - text-retrieval

EMS-Knowledge

EMS-Knowledge is a lightweight, retrieval-friendly dataset distilled from Emergency Medical Services (EMS) reference materials.
Each row contains a short section of text paired with its file name, topic, and section title, making it easy to build search / RAG pipelines, create study aids, or ground QA systems. See more on our project page.

This repository includes:

A tabular split (single split named train) with the following columns:
- file — source filename (e.g., anatomy.json)
- source — folder label (here: knowledge)
- topic — coarse topic derived from file stem (e.g., anatomy, trauma)
- section — section/heading within the file
- text — the normalized section text
The original JSON files under raw/ for direct use:
- airway, respiratory, ventilation.json
- anatomy.json
- assessment.json
- cardiovascular.json
- ems_operations.json
- medical.json
- others.json
- pediatrics.json
- pharmacology.json
- terminology.json
- trauma.json

💡 The JSON files are heterogeneous collections of (section → text) entries. The table view flattens these into uniform rows for easier indexing and filtering.

Dataset Structure

Split: train (tabular)

Columns

file (string)
source (string)
topic (string)
section (string)
text (string)

Cardinality: depends on the number of sections parsed from the JSONs.
(Each JSON entry becomes one row.)

Citation

If you use EMS-MCQA in your work, please cite:

@misc{ge2025expertguidedpromptingretrievalaugmentedgeneration,
  title         = {Expert-Guided Prompting and Retrieval-Augmented Generation for Emergency Medical Service Question Answering},
  author        = {Xueren Ge and Sahil Murtaza and Anthony Cortez and Homa Alemzadeh},
  year          = {2025},
  eprint        = {2511.10900},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CL},
  url           = {https://arxiv.org/abs/2511.10900},
}

Quick Start

from datasets import load_dataset

# Load table (single split named "train")
ds = load_dataset("Xueren/EMS-Knowledge", split="train")
print(ds)
print(ds[0])

# Filter by topic
anatomy = ds.filter(lambda x: x["topic"] == "anatomy")

# Simple keyword search over section text
needle = "spinal cord"
hits = ds.filter(lambda x: needle.lower() in x["text"].lower())
print(len(hits), "matches for", needle)

# Group sections by file/topic (example)
by_file = ds.to_pandas().groupby("file").size().sort_values(ascending=False)
print(by_file.head())