dual-use-papers / README.md
AE-data's picture
Upload README.md with huggingface_hub
682f04a verified
---
license: cc-by-4.0
task_categories:
- text-classification
- text-generation
language:
- en
tags:
- academic-papers
- virology
- nuclear-physics
- cybersecurity
- astrophysics
- grobid
size_categories:
- 100K<n<1M
---
# Dual-Use Papers Dataset
A large-scale collection of academic papers converted to Markdown, spanning multiple scientific domains relevant to dual-use research.
## Dataset Description
This dataset contains ~360k+ full-text academic papers sourced from Europe PMC, arXiv, and DOE OSTI, processed into clean Markdown using [GROBID](https://github.com/kermitt2/grobid).
### Columns
| Column | Type | Description |
|--------|------|-------------|
| `category` | `string` | High-level topic: `biology`, `nuclear`, `cyber`, `astrophysics` |
| `source` | `string` | Specific origin (see below) |
| `url` | `string` | URL of the source PDF |
| `text` | `string` | Full paper text in Markdown format |
### Sources
| Source | Category | Description |
|--------|----------|-------------|
| `europe-pmc` | biology | Virology papers from Europe PMC (1990–2026) |
| `osti` | nuclear | DOE technical reports, journal articles, books, theses |
| `arxiv_cs.CR` | cyber | Cryptography and Security |
| `arxiv_cs.NI` | cyber | Networking and Internet Architecture |
| `arxiv_cs.OS` | cyber | Operating Systems |
| `arxiv_nucl-th` | nuclear | Nuclear Theory |
| `arxiv_nucl-ex` | nuclear | Nuclear Experiment |
| `arxiv_astro-ph` | astrophysics | Astrophysics (all subcategories, 2020+) |
The category `astrophysics` is included as a control to provide papers of a similar format as the rest, without being dual use.
### Data Collection
- **Europe PMC**: Papers matching `Virology AND (HAS_FT:Y) AND (FIRST_PDATE:[1990 TO 2026])` via the Europe PMC REST API.
- **arXiv**: Papers from specified categories via the arXiv search API, paginated by year windows.
- **OSTI**: Full-text papers from DOE OSTI.gov matching semantic query "weapon" across Journal Articles, Technical Reports, Books, Program Documents, and Theses/Dissertations.
### Processing Pipeline
1. **Metadata collection** — PDF URLs and metadata gathered from each source's API
2. **PDF download + GROBID conversion** — PDFs downloaded ephemerally, converted to TEI XML via GROBID, then transformed to Markdown
3. **Dataset assembly** — Markdown files joined with metadata, filtered to only include successfully converted papers
### Text Format
Each `text` field contains structured Markdown with:
- Paper title (H1)
- Author list
- Abstract (H2)
- Body sections with headings
- Figures/tables (as blockquotes)
- Formulas (LaTeX)
- Numbered references
## Licensing
This dataset is distributed under CC-BY-4.0 for the dataset structure and metadata. However, each individual text inherits the license of its original source publication. Users should consult the original source (via the `url` field) to determine the specific license terms for any given paper before redistribution or commercial use.
## Limitations
- Text quality depends on GROBID's ability to parse each PDF; some papers (especially scanned/image-heavy ones) may have lower quality output.
- Not all papers from each source have successfully converted Markdown — only those with non-empty conversions are included.
- arXiv astrophysics papers are limited to 2020 onwards.