Hugging Face's logo

Datasets:
bluuebunny
/
crossref_metadata_2025_split

Tasks:
Sentence Similarity
Modalities:
Text
Formats:
parquet
Languages:
English
Size:
10M - 100M
Tags:
doi
bibliography
literature
crossref
Libraries:
Datasets
Dask
License:
Dataset card Data Studio Files
xet
Community
crossref_metadata_2025_split / README.md
bluuebunny's picture
bluuebunny
Update README.md
a6a5f25 verified
preview code
|
raw
history blame contribute delete
4.16 kB
metadata
license: apache-2.0
task_categories:
  - sentence-similarity
language:
  - en
tags:
  - doi
  - bibliography
  - literature
  - crossref
pretty_name: crossref 2025
size_categories:
  - 10M<n<100M

Dataset Overview

This dataset contains bibliographic metadata from the public Crossref snapshot released in 2025. It provides core fields for scholarly documents, including DOI, title, abstract, authorship, publication month and year, and URLs. The entire public dump (~196.94 GB) was filtered and extracted into a parquet format for efficient loading and querying.

  • Total size: 196.94 GB (parquet files)
  • Number of records: 34,308,730

Use this dataset for large-scale text mining, bibliometric analyses, metadata enrichment, and building citation-aware tools.

Dataset Features

Each record in the dataset contains the following fields:

Field Type Description
doi string Digital Object Identifier of the publication.
title string Title of the scholarly work.
abstract string Abstract text (when available).
author list List of author names or structured author metadata.
month int Publication month (1–12).
year int Publication year (e.g., 2024, 2025).
url string URL pointing to the publication page or DOI resolver.

Dataset Structure

The dataset is provided in Apache Parquet format. Parquet allows for efficient columnar storage and supports schema evolution. Each parquet file chunk contains the complete schema as described above.

root
 |-- doi: string (nullable = true)
 |-- title: string (nullable = true)
 |-- abstract: string (nullable = true)
 |-- author: array (nullable = true)
 |    |-- element: string (nullable = true)
 |-- month: int (nullable = true)
 |-- year: int (nullable = true)
 |-- url: string (nullable = true)

Dataset Splits

This dataset does not come with predefined splits. Users can split based on publication year, subject areas, or random sampling as per their experimentation needs.

Dataset Creation

Source

Processing

  1. Extraction: Parsed the Crossref dump to extract relevant fields (DOI, title, abstract, authors, month, year, URL).
  2. Transformation: Normalized fields; authors consolidated into a list of names.
  3. Serialization: Saved the resulting table in Parquet format for columnar efficiency.
  4. Storage: Uploaded parquet files to Hugging Face Datasets with corresponding metadata.

Code for dataset processing and card generation is available at:

https://github.com/mitanshu7/PaperMatch_crossref

Usage 

from datasets import load_dataset

dataset = load_dataset(
    "bluuebunny/crossref_metadata_2025",
    streaming=True,
    split='train'
)

# Inspect a few records
print(dataset[0])

# Filter by year
subset_2025 = dataset.filter(lambda x: x["year"] == 2025)
print(f"Records published in 2025: {len(subset_2025)}")

Citation

If you use this dataset in your research, please cite the Crossref public data file:

@misc{crossref2025,
  title        = {{Crossref} Public Data File 2025},
  author       = {{Crossref}},
  year         = 2025,
  howpublished = {\url{https://www.crossref.org/blog/2025-public-data-file-now-available/}},
}

Contact