---
language:
  - en
task_categories:
  - text-mining
  - other
tags:
  - scientific-papers
  - grobid
  - tei-xml
  - bibtex
pretty_name: "ICLR/ICML/NeurIPS 2025 (no-PDF) GROBID outputs"
license: other
---

## Dataset Summary

This dataset contains **GROBID-parsed outputs** for research papers from **ICLR 2025**, **ICML 2025**, and **NeurIPS 2025**.  
The repository is distributed as **zip archives** (no PDFs) to make it easy to download and mirror.

What you will find (per paper, when available):
- **TEI XML** produced by GROBID (`*.xml`)
- **BibTeX** produced by GROBID (`*.bib`)
- In some folders: additional GROBID artifacts such as `grobid_metadata` JSON and `grobid_bib` exports

What you will *not* find in the current release:
- PDFs (explicitly excluded)
- OpenReview reviews / scores
- A single tabular `datasets`-style split (this repo is file-based)

## Dataset Structure

The dataset is organized by conference year folders and typically shipped as zips:

```
paper_data/
├── ICLR_2025_no_pdf.zip
├── ICML_2025_no_pdf.zip
├── NeurIPS_2025_no_pdf.zip
```

Inside each zip (example; exact subfolders can differ by venue):

```
<VENUE>_2025/
├── grobid_tei/          # TEI XML files (*.xml)
├── grobid_bib/          # BibTeX exports (*.bib) (venue-dependent)
├── grobid_metadata/     # JSON metadata (venue-dependent)
└── ...                  # other non-PDF artifacts
```

| Conference   | Papers | Reviews |
|------------|------:|-------:|
| ICLR 2025 | 11,475 | 46,748 |
| ICML 2025 | 3,385 | 35,546 |
| NeurIPS 2025 | 5,532 | 22,373 |

## How to Use

### Download from Hugging Face Hub

```python
from huggingface_hub import snapshot_download

local_dir = snapshot_download(repo_id="thanhkt/paper", repo_type="dataset")
print(local_dir)
```

Then unzip the archives you need (example):

```bash
unzip -q ICLR_2025_no_pdf.zip -d ./extracted/
```

### Parsing

- **TEI XML**: use any TEI/XML parser to extract title/abstract/sections/citations.
- **BibTeX**: parse with `bibtexparser` or similar libraries.

## Limitations

- GROBID outputs may contain parsing errors and incomplete fields depending on paper formatting.
- File coverage varies by venue and crawling/processing completeness.
- Because this repo is primarily **zip/binary files**, the Hugging Face dataset viewer may not display a table preview.