yufan/arxiv-metadata-2020-2026
arXiv Metadata, enriched (2020–2026) Per-paper metadata for 1,517,185 arXiv papers spanning 2020-01 → 2026-09, enriched with abstracts, citation counts, and Semantic Scholar identifiers, and organized as a two-level hierarchy: field of study → year. Unlike a bare title index, every record here carries the abstract, the full author list, citation counts, and the Semantic Scholar corpusId, so you can do retrieval, classification, citation analysis, and corpus building directly… See the full description on the dataset page: https://huggingface.co/datasets/yufan/arxiv-metadata-2020-2026.
arXiv Metadata, enriched (2020–2026)
Per-paper metadata for 1,517,185 arXiv papers spanning 2020-01 → 2026-09, enriched with abstracts, citation counts, and Semantic Scholar identifiers, and organized as a two-level hierarchy: field of study → year.
Unlike a bare title index, every record here carries the abstract, the full author list, citation counts, and the Semantic Scholar `corpusId`, so you can do retrieval, classification, citation analysis, and corpus building directly from the metadata.
Layout: field → year
Files are stored as `<Field>/<Year>/metadata.jsonl`, one JSON object per line. Fields (the first level) are ordered largest → smallest by paper count:
(– means that field has no papers harvested for that year.)
Columns
Usage
from datasets import load_dataset
# One field, all years (each year is a split):
cs = load_dataset("yufan/arxiv-metadata-2020-2026", "Computer_Science")
print(cs) # splits: 2020 … 2026
print(cs["2025"][0])
# One field + one year:
phys25 = load_dataset("yufan/arxiv-metadata-2020-2026", "Physics", split="2025")
# Stream the largest split without downloading everything:
it = load_dataset("yufan/arxiv-metadata-2020-2026", "Computer_Science",
split="2025", streaming=True)
print(next(iter(it)))Download a single raw file directly:
from huggingface_hub import hf_hub_download
path = hf_hub_download(
repo_id="yufan/arxiv-metadata-2020-2026",
filename="Computer_Science/2025/metadata.jsonl",
repo_type="dataset",
)Provenance
- arXiv — base identifiers and titles, via arXiv's open metadata.
- Semantic Scholar — abstracts, author lists,
corpusId,fieldsOfStudy, and citation counts, via the Semantic Scholar Academic Graph API. - Date window: 2020-01 through ~July 2026; 2026 is partial.
fieldsOfStudyis Semantic Scholar's primary field; papers it could not place are grouped under `Unknown`.
License & attribution
The enriched fields originate from the Semantic Scholar Academic Graph, released under ODC-BY 1.0 — please attribute Semantic Scholar (and arXiv for the underlying papers) when you use this data. arXiv's own metadata is provided under CC0; see arXiv's Terms of Use.
Citation
@misc{arxiv_metadata_enriched_2020_2026,
title = {arXiv Metadata, enriched (2020--2026), by field and year},
author = {yufan},
year = {2026},
url = {https://huggingface.co/datasets/yufan/arxiv-metadata-2020-2026}
}