README.md

---
license: cc-by-4.0
task_categories:
- graph-ml
language:
- en
pretty_name: Compact OpenAIRE Citation Graph
size_categories:
- 1B<n<10B
tags:
- citation-graph
- citation-network
- scholarly-data
- bibliometrics
- scientometrics
- openaire
- graph
- dynamic-graph
- temporal-graph
- dynamic-network
- link-prediction
configs:
- config_name: citations
  data_files: citations.parquet
- config_name: publications_large
  data_files: publications_large.parquet
---

<div style="display: flex; gap: 8px; align-items: center; flex-wrap: wrap; margin-bottom: 16px;">
  <a href="https://doi.org/10.5334/johd.520"><img src="https://img.shields.io/badge/Paper-JOHD%2012(1)%2063-blue" alt="Paper"></a>
  <a href="https://doi.org/10.5281/zenodo.18402099"><img src="https://img.shields.io/badge/Zenodo-10.5281%2Fzenodo.18402099-blue" alt="Zenodo DOI"></a>
  <a href="https://graph.openaire.eu/docs/"><img src="https://img.shields.io/badge/docs-OpenAIRE%20Graph-informational" alt="OpenAIRE Graph Docs"></a>
  <img src="https://img.shields.io/badge/license-CC--BY--4.0-green" alt="License">
</div>

# 📚 Compact OpenAIRE Citation Graph

<!-- BASED-ON:START -->
*Based on **OpenAIRE Graph v11.1.1** ([source on Zenodo](https://zenodo.org/records/20428976)).*
<!-- BASED-ON:END -->

> The complete OpenAIRE citation graph, distilled into a handful of compact, analysis-ready files — the full scholarly citation network of the open-science ecosystem, small enough to actually work with.

Citation graphs at this scale are usually locked behind multi-terabyte dumps and heavyweight infrastructure. This dataset makes the **entire OpenAIRE citation network** loadable on a normal machine: publications as nodes, citations as edges, served as compressed Parquet with a memory-efficient loading path. A richer node file adds titles, abstracts, authors, dates, and a full set of persistent identifiers (DOI, PubMed, MAG, arXiv, and more) for text-attributed and metadata-driven work.

**Please cite the paper if you use this data.**

### At a glance

| | |
| --- | --- |
| **Nodes** | Scholarly publications |
| **Edges** | Citation links |
| **Format** | Parquet (PyArrow-friendly) |
| **Node metadata** | Titles, abstracts, authors, dates, venues, PIDs |
| **Best for** | Graph ML · temporal / dynamic graphs · text-attributed graphs · bibliometrics |
| **License** | CC-BY-4.0 |
| **Cite** | [Skarding & Sanda (2026), *JOHD* 12(1), 63](https://doi.org/10.5334/johd.520) |
| **Dataset DOI** | [10.5281/zenodo.18402099](https://doi.org/10.5281/zenodo.18402099) |

## Dataset structure

The dataset is a directed citation graph: publications are nodes, citations are edges.

| File | Role | Size (Parquet) | Number of entries |
| --- | --- | --- | --- |
| `citations.parquet` | Edges — the citation links between publications | 9.2 GB | ~2.37B |
| `publications_large.parquet` | Nodes with additional metadata fields (see below) | 75.5 GB | ~216M |

### Features/columns in `publications_large`

| Field | Type | Description | Memory (GB) | Filled |
| --- | --- | --- | --- | --- |
| `nodeId` | int32 | Unique internal identifier for the node (publication) | 0.8 | 100.00% |
| `openaireId` | str | Identifier assigned by the OpenAIRE platform | 10.1 | 100.00% |
| `title` | str | Title of the publication | 17.3 | 99.40% |
| `authors` | list[str] | List of authors associated with the publication | 11.6 | 83.78% |
| `description` | str | Abstract or short description of the publication | 137.6 | 57.10% |
| `date` | datetime | Date when the publication was published | 0.8 | 97.33% |
| `container` | str | Journal, conference, or repository where it was published | 5.7 | 68.35% |
| `citations` | int | Number of times the publication has been cited | 1.6 | 97.62% |
| `language` | str | Language in which the publication is written | 1.5 | 100.00% |
| `pid_dois` | list[str] | DOI identifiers | 5.9 | 80.60% |
| `pid_mag_ids` | list[str] | MAG IDs | 2.0 | 39.50% |
| `pid_pmids` | list[str] | PubMed IDs | 1.3 | 18.14% |
| `pid_handles` | list[str] | Persistent handles | 1.2 | 8.35% |
| `pid_pmcs` | list[str] | PubMed Central IDs | 1.0 | 4.78% |
| `pid_arxiv_ids` | list[str] | ArXiv IDs | 0.9 | 1.38% |

## Quickstart
```python
import pandas as pd
from huggingface_hub import hf_hub_download

REPO = "Zmeos/Compact_OpenAIRE_citation_graph"

# citations (edges)
cites = pd.read_parquet(
    hf_hub_download(REPO, "citations.parquet", repo_type="dataset"),
    engine="pyarrow", dtype_backend="pyarrow",
)

# publications_large (nodes + metadata) — may not fit in memory;
# select only the columns you need
large = pd.read_parquet(
    hf_hub_download(REPO, "publications_large.parquet", repo_type="dataset"),
    columns=["nodeId", "title", "pid_dois"],
    engine="pyarrow", dtype_backend="pyarrow",
)
```

## Reproducibility

The PySpark pipeline used to produce these files (with a Singularity/Apptainer container
for portability) is archived on Zenodo as `pipeline.tar.xz` and maintained at
[Codeberg](https://codeberg.org/Zmeos/OpenAIRE-citation-extraction).

## License

Creative Commons Attribution 4.0 International (CC-BY-4.0).

## Citation (paper)

When using this dataset, please cite the accompanying article:

> Skarding, J. and Sanda, P. (2026) 'Making the Complete OpenAIRE Citation Graph Easily
> Accessible Through Compact Data Representation', *Journal of Open Humanities Data*,
> 12(1), p. 63. https://doi.org/10.5334/johd.520

```bibtex
@article{skarding2026openaire,
  author  = {Skarding, Joakim and Sanda, Pavel},
  title   = {Making the Complete OpenAIRE Citation Graph Easily Accessible Through Compact Data Representation},
  journal = {Journal of Open Humanities Data},
  volume  = {12},
  number  = {1},
  pages   = {63},
  year    = {2026},
  doi     = {10.5334/johd.520}
}
```

Dataset archive: Skarding, J. and Sanda, P. (2026). *Compact representation of the
OpenAIRE citation graph* [Data set]. Zenodo.
https://doi.org/10.5281/zenodo.18402099
---