---
license: cc0-1.0
language:
- en
tags:
- pubmed
pretty_name: PubMedAbstractSubset
---
# PubMed Abstracts Subset



This dataset contains a probabilistic sample of publicly available PubMed metadata sourced from the [National Library of Medicine (NLM)](https://pubmed.ncbi.nlm.nih.gov/).

If you're looking for the precomputed embedding vectors (MedCPT) used in our work [*Efficient and Reproducible Biomedical Question Answering using Retrieval Augmented Generation*](https://arxiv.org/abs/2505.07917), they are available in a separate dataset: [slinusc/PubMedAbstractsSubsetEmbedded](https://huggingface.co/datasets/slinusc/PubMedAbstractsSubsetEmbedded).

---

## 📄 Description

Each entry in the dataset includes:

- `title`: Title of the publication
- `abstract`: Abstract text
- `PMID`: PubMed identifier

The dataset is split into 24 `.jsonl` files, each containing approximately 100,000 entries, for a total of ~2.39 million samples.

---

## 🔍 How to Access

### ▶️ Option 1: Load using Hugging Face `datasets` (streaming)

```python
from datasets import load_dataset

dataset = load_dataset("slinusc/PubMedAbstractsSubset", streaming=True)

for doc in dataset:
    print(doc["title"], doc["abstract"])
    break
```

> Streaming is recommended for large-scale processing and avoids loading the entire dataset into memory.

---

### 💾 Option 2: Clone using Git and Git LFS

```bash
git lfs install
git clone https://huggingface.co/datasets/slinusc/PubMedAbstractsSubset
cd PubMedAbstractsSubset
```

> After cloning, run `git lfs pull` if needed to retrieve the full data files.

---

## 📦 Format

Each file is in `.jsonl` (JSON Lines) format, where each line is a valid JSON object:

```json
{
  "title": "...",
  "abstract": "...",
  "PMID": 36464820
}
```

---

## 📚 Source and Licensing

This dataset is derived from public domain PubMed metadata (titles and abstracts), redistributed in accordance with [NLM data usage policies](https://www.nlm.nih.gov/databases/download/data_distrib_main.html).
  
- Used in:  
  **Stuhlmann et al. (2025)**, *Efficient and Reproducible Biomedical QA using RAG*, [arXiv:2505.07917](https://arxiv.org/abs/2505.07917)

  https://github.com/slinusc/medical_RAG_system

---

## 🏷️ Version

- `v1.0` – Initial release (2.39M entries, 24 JSONL files)

---

## 📬 Contact

Maintained by [@slinusc](https://huggingface.co/slinusc).  
For questions or issues, please open a discussion or pull request on the Hugging Face dataset page.