ashesborn6/retracted-papers-citation-integrity-graph

Retracted Papers: Citation and Retraction Integrity Graph

A citation and retraction-propagation graph over scholarly works, for research-integrity and scientometrics work: retracted papers, the works that cite them, and source-attributed retraction status facts. Use it to study how a retraction propagates through a citation neighbourhood, to audit whether a reference list cites retracted or paper-mill-adjacent work, and to ground a retraction-status answer in inspectable records instead of a model's memory.

Not a compliance or legal determination. The dataset reports source-attributed facts; turning them into a verdict is the job of the endpoint described below.

Live demo: ashesborn6/retraction-status looks up any DOI and shows the rows behind the answer.

Quick start

from datasets import load_dataset

papers = load_dataset("ashesborn6/integrity-citation-retraction-graph", "papers", split="train")
cites = load_dataset("ashesborn6/integrity-citation-retraction-graph", "cites", split="train")
status = load_dataset("ashesborn6/integrity-citation-retraction-graph", "retraction_status", split="train")

retracted = papers.filter(lambda r: r["openalex_is_retracted"])
print(len(papers), len(cites), len(status), len(retracted))  # 98868 144814 572 197

Without the datasets library:

base=https://huggingface.co/datasets/ashesborn6/integrity-citation-retraction-graph/resolve/main/data
for f in papers cites retraction_status; do curl -sL "$base/$f.jsonl.gz" | gzip -dc > "$f.jsonl"; done
wc -l papers.jsonl cites.jsonl retraction_status.jsonl

One DOI, by hand:

curl -s "https://pay.ashesborn.cloud/api/v1/judge/paper?doi=10.1016/S0140-6736(97)11096-0"

What is in the release

Config	Rows	Columns
papers	98,868 (96,117 distinct DOIs, publication years 1743–2026)	paper_id, doi, title, publication_year, cited_by_count, openalex_is_retracted
cites	144,814 citation edges (98,329 distinct cited works)	citing_id, cited_id, cited_doi, source_tier, source_uri
retraction_status	572 status facts over 235 papers	paper_id, doi, source, source_tier, is_retracted, retraction_date

openalex_is_retracted is true for 197 papers. The 572 status rows split 235 OpenAlex / 188 Retraction Watch / 149 Crossref, and 453 of them assert is_retracted: true across 197 distinct DOIs, so a single retracted paper usually carries three independently sourced assertions. Every citation edge in this release is source_tier: C (OpenAlex); no other citation source is included yet.

How it was built

The snapshot was assembled from the OpenAlex entity API, Crossref update-to metadata, and Retraction Watch status facts only. Retraction Watch free-text reasons and nature descriptions are excluded from this export.

Provenance

• data/papers.jsonl.gz: OpenAlex contributes paper_id, doi, title, publication_year, cited_by_count, and openalex_is_retracted.
• data/cites.jsonl.gz: OpenAlex/OpenCitations citation records contribute citing_id, cited_id, and source_uri; the cited work's OpenAlex record contributes cited_doi; source_tier records the source classification.
• data/retraction_status.jsonl.gz: OpenAlex supplies paper_id, doi, and is_retracted in OpenAlex status rows; Crossref supplies is_retracted and retraction_date in Crossref status rows; Retraction Watch supplies is_retracted and retraction_date in Retraction Watch status rows. source and source_tier label each source-attributed row.

Rows are JSON Lines with deterministic key order. The status file deliberately contains facts only; it does not redistribute Retraction Watch narrative text.

Limits

• This is a neighbourhood sample: 12 retracted seeds with up to 60 citing works per seed, expanded through citation links. It is not a full scholarly corpus, and it is not a retracted-paper registry. 197 retracted papers is a small fraction of the ~70,000 notices recorded by Retraction Watch and Crossref combined.
• Coverage varies by paper and by how much source metadata exists. Semantic Scholar citation contexts are intentionally absent.
• retraction_date is null in the OpenAlex status rows; Crossref and Retraction Watch rows carry the date. That discrepancy is part of the data, not a gap to fill in.
• A graph edge or status fact is evidence to inspect, not by itself a conclusion about the merits of a claim.

How to reproduce

Run the exporter from integrations/mcp against a harvested snapshot:

bun scripts/integrity-export-hf.ts --snapshot <snapshotDir> --out <dir>

The output directory contains the three compressed JSONL files, this dataset card, attribution, and a manifest with hashes, byte sizes, and row counts.

How to verify a claim

For a claim-specific comparison, consult the judgment API at https://pay.ashesborn.cloud and compare the cited work identifiers and status facts with the corresponding source records. To check one row by hand, open the DOI at https://doi.org/<doi>, compare retraction_date against the publisher notice, and compare source/source_tier across the three status rows for that paper.

Citation

@misc{integrity_citation_retraction_graph,
  title = {Retracted Papers and Citation Integrity Graph},
  year = {2026},
  howpublished = {Hugging Face dataset},
  note = {Snapshot ri-20261025-1526; data/papers.jsonl.gz sha256 87adac4fcdec1c41d612595236dd08c40565d9cd44aae937c6f21884055954c8}
}

中文摘要

本数据集提供学术论文元数据、引用关系及撤稿状态事实，用于研究撤稿（retracted papers）在引用网络中的传播， 也可用于检测参考文献是否引用了已撤稿或论文工厂（paper mill）相关的成果。数据源为 OpenAlex、Crossref 与 Retraction Watch，仅包含事实，不含 Retraction Watch 的原因文本。数据只覆盖有限的撤稿论文邻域 （197 篇撤稿论文、144,814 条引用边），不是完整的撤稿论文库。在线演示： https://huggingface.co/spaces/ashesborn6/retraction-status