CoChat

Identifying fabricated networks within authorship-for-sale enterprises

Open access
Porter, Simon J., McIntosh, Leslie D. · 2024 · Scientific Reports

The DOI matches its Crossref record; 37 of the 37 references it cites were found in a registry or an index; 5 of the 43 sources its claims cite were checked for support.

Abstract

It is estimated that 2% of all journal submissions across all disciplines originate from paper mills, both creating significant risk that the body of research that we rely on to progress becomes corrupted, and placing undue burden on the submission process to reject these articles. By understanding how the business of paper mills—the technological approaches that they adopt, as well as the social structures that they require to operate—the research community can be empowered to develop strategies that make it harder, or ideally impossible for them to operate. Most of the contemporary work in paper-mill detection has focused on identifying the signals that have been left behind inside the text or structure of fabricated papers that result from the technological approaches that paper mills employ. As technologies employed by paper mills advance, these signals will become harder to detect. Fabricated papers do not just need text, images, and data however, they also require a fabricated or partially fabricated network of authors. Most ‘authors’ on a fabricated paper have not been associated with the research, but rather are added through a transaction. This lack of deeper connection means that there is a low likelihood that co-authors on fabricated papers will ever appear together on the same paper more than once. This paper constructs a model that encodes some of the key characteristics of this activity in an ‘authorship-for-sale’ network with the aim to create a robust method to detect this type of activity. A characteristic network fingerprint arises from this model that provides a robust statistical approach to the detection of paper-mill networks. The model suggested in this paper detects networks that have a statistically significant overlap with other approaches that principally rely on textual analysis for the detection of fraudulent papers. Researchers connected to networks identified using the methodology outlined in this paper are shown to be connected with 37% of papers identified through the tortured-phrase and clay-feet methods deployed in the Problematic Paper Screener website. Finally, methods to limit the expansion and propagation of these networks is discussed both in technological and social terms.

Overview

CoChat Summary

TL;DR

The paper models authorship-for-sale paper mills as fabricated co-authorship networks and shows that unusual network shapes—young publication age, low egocentric clustering, high annual output, and a connected component—can flag likely paper-mill activity. It validates the network signal against the Problematic Paper Screener and Retraction Watch, finding substantial overlap and low false-positive rates in random samples.

Key findings

  • In 2022, only 0.10% of Stage I and 0.43% of Stage II researchers had network shapes repeated fewer than 10 times; the unusual-author cohort was 0.020% of Stage I/II researchers before 2018, doubling over four years from 2018, with the largest connected component rising from 0.002% of authors (2010–2017) to 0.034% in 2022 (about 75% of unusual authors).
  • For 2020–2022, 2,056 researchers met the unusual-author criteria. Of 3,739 Problematic Paper Screener publications, 10% (367) included a researcher in the authorship-for-sale profile and 37% (1,371) included a co-author of one; 72% of the cohort connected back to the tortured-phrase dataset versus 16% direct authors. Random samples had 1.2% direct and 16% collaborator overlap.
  • Unusual authors appeared on 7.43% of 1,858 Retraction Watch research articles (2020–2022), over six times the random rate. Among 741 Stage I/II researchers with >250 ORCID peer reviews, 8.2% were in the cohort and 51% had co-authored with someone in it, representing 218,907 claimed reviews.
  • By 2022, Elsevier had >11,000 implicated publications, MDPI just under 7,000, and Springer Nature over 5,000; Hindawi’s risk profile doubled to 4%, MDPI climbed to just under 3%, and most publishers sustained 1–2%. 12,434 journals had <2% exposure, 696 had 2–4%, and 717 had >4%. Saudi Arabia had 1.2% of its Stage I–II workforce identified, corresponding to 22.6% of research articles and reviews.

Methods

  • Used Dimensions (Digital Science) data on Google BigQuery, with researcher disambiguation, ORCID, Retraction Watch, and Problematic Paper Screener tortured-phrase/clay-feet datasets. Defined publication age and network shape; clustering coefficient C = E / possible edges after removing the central node.
  • Filtered to research articles, excluded >20 authors, assigned career stages in 5-year publication-age increments, restricted to >20 publications in the same calendar year, most frequent collaborator Stage I–III, >50% network Stage I–II, and largest connected component per publication year plus following year; graph calculations used NetworkX.
  • Validated against PPS tortured-phrase/clay-feet papers, Retraction Watch, ORCID high-volume peer reviewers, and Monte Carlo random samples matched in size and period.

Limitations

  • The paper states author disambiguation is hardest for young researchers and some writing systems, with false positives unavoidable due to merged or split profiles; all flagged authors need human curation, and ORCID adoption is recommended. It also states the method only identifies statistical outliers, not proof, and individual investigation is required.
  • The paper states it only detects “outside-the-tent” misconduct; inside-the-tent image manipulation, plagiarism, data fabrication, or purchasing entire papers will not be detected. It also notes exclusions (>20 authors, single publication year) may miss activity, and brokered authorship-for-sale connects to organic networks so not all connected researchers are culpable.
  • Evident from the text: reliance on Dimensions disambiguation and ORCID/PPS data quality; PPS false positives remain and papers with <5 tortured phrases are not public; ORCID peer-review counts may double count; GDPR limits direct data sharing.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41598-024-71230-8

Cites

37 of 37 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 Histogram of Stage I and Stage II researchers with network shapes that have been been repeated fewer than 10 times in 2022. Researchers are distributed across the x-axis by their clustering coefficient. The middle green histogram represents the subset of researchers that have produced greater than 20 publications in the year. In the bottom red subset, the most frequent collaborator is a Stage I-III researcher, and greater that 50% of the network is made up of Stage I-II researchers.
Fig. 1 Histogram of Stage I and Stage II researchers with network shapes that have been been repeated fewer than 10 times in 2022. Researchers are distributed across the x-axis by their clustering coefficient. The middle green histogram represents the subset of researchers that have produced greater than 20 publications in the year. In the bottom red subset, the most frequent collaborator is a Stage I-III researcher, and greater that 50% of the network is made up of Stage I-II researchers.CC BY · PubMed Central
Fig. 2 Development of the unusual author cohort as a percentage of all Stage I and II researchers by year. The two areas are overlapping and not cumulative. The area between the top of the light blue region and the x-axis represents all unusual Stage I and II researchers as a percentage of all Stage I and Stage II researchers. The dark blue area shows just those that are part of the largest connected component of the unusual co-author network defined by our model.
Fig. 2 Development of the unusual author cohort as a percentage of all Stage I and II researchers by year. The two areas are overlapping and not cumulative. The area between the top of the light blue region and the x-axis represents all unusual Stage I and II researchers as a percentage of all Stage I and Stage II researchers. The dark blue area shows just those that are part of the largest connected component of the unusual co-author network defined by our model.CC BY · PubMed Central
Fig. 3 Histogram of the graph density of the co authorship networks of random samples of researchers with clustering coefficients of less than 0.4 and greater than 20 publications. The sample size of each measurement is the same size as the unusual author sample.
Fig. 3 Histogram of the graph density of the co authorship networks of random samples of researchers with clustering coefficients of less than 0.4 and greater than 20 publications. The sample size of each measurement is the same size as the unusual author sample.CC BY · PubMed Central
Fig. 4 Proportion of papers exhibiting tortured phases as a percentage of all journal articles produced in a year. The tortured-phases dataset was extracted from the Problematic Paper Screener. The drop in papers in 2022 might be explained by the implementation of the tortured phrases detection strategies into submission processing workflows, or a reaction to detection by paper mills.
Fig. 4 Proportion of papers exhibiting tortured phases as a percentage of all journal articles produced in a year. The tortured-phases dataset was extracted from the Problematic Paper Screener. The drop in papers in 2022 might be explained by the implementation of the tortured phrases detection strategies into submission processing workflows, or a reaction to detection by paper mills.CC BY · PubMed Central
Fig. 5 Publications by volume that have an author in the unusual author set.
Fig. 5 Publications by volume that have an author in the unusual author set.CC BY · PubMed Central
Fig. 6 Risk profile of exposure to the unusual author cohort. The percentage of publications exposed is expressed as a percentage of publisher output.
Fig. 6 Risk profile of exposure to the unusual author cohort. The percentage of publications exposed is expressed as a percentage of publisher output.CC BY · PubMed Central
Fig. 7 Histogram of journal exposure unusual author cohort. The x-axis is the percentage of papers in any given journal that appear to be associated with the unusual paper network (i.e. number of papers found in a journal that appear to be authored by a unusual author). The y-axis, using a log10-scale, is the frequency with which journals fall into each 1% band of exposure. There are 12,434 journals (shown in green) that contain potentially unusual-author-affiliated publications at a rate between 0 and 2%; 696 journals (shown in orange) have an exposure of between 2% and 4% of their articles; 717 journals have an exposure to the unusual-author network with more than 4% of their articles. Note that the log10 scale of the y-axis compresses and underplays the number of “safe” journals and tends to emphasize the size of the tale of this distribution.
Fig. 7 Histogram of journal exposure unusual author cohort. The x-axis is the percentage of papers in any given journal that appear to be associated with the unusual paper network (i.e. number of papers found in a journal that appear to be authored by a unusual author). The y-axis, using a log10-scale, is the frequency with which journals fall into each 1% band of exposure. There are 12,434 journals (shown in green) that contain potentially unusual-author-affiliated publications at a rate between 0 and 2%; 696 journals (shown in orange) have an exposure of between 2% and 4% of their articles; 717 journals have an exposure to the unusual-author network with more than 4% of their articles. Note that the log10 scale of the y-axis compresses and underplays the number of “safe” journals and tends to emphasize the size of the tale of this distribution.CC BY · PubMed Central

Verification

Claims the paper makes

3 Valid · 7 Not valid · 30 Not checked · 201 Own results and interpretations

96 more claims

Create a free account to see all 241 claims and to verify the 38 references still open.

See all claims — free account
  • Probably not a substantial claim. 100
  • It may state two separate claims as one. 64
  • The quote only partly states this claim. 1

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

4 of 4

Citations

Cites

1 retracted

Show all 37 references

Research works printed but not linked: 7

  1. Chawla, D. S. Exposing a most unscrupulous journal (2023).
  2. Barbour, B. & Stell, B. M. Gaming the Metrics (The MIT Press, London, 2020).
  3. Montenegro, A. ORCID Public Data File 2023. Tech. Rep., ORCID (2023). https://doi.org/10.23640/07243.24204912.v1. · DOI
  4. {Millman”], [”Gaël Varoquaux and Travis Vaught and Jarrod} (ed.). Exploring Network Structure, Dynamics, and Function using NetworkX (2008).
  5. FLYNN, J. Guest Post-Addressing Paper Mills and a Way Forward for Journal Security.
  6. Flintoft, L., MacCallum, S. P. D. a. W. C. J., Streeter, D. o. F. S. R. a. W. M., Flanagan, Director of Research Integrity Strategy and Policy at Wiley Dave & Ferguson, S. D. o. D. S. a. W. L. Tackling publication manipulation at scale: Hindawi’s journey and lessons for academic publishing. Tech. Rep., Hindawi (2023).
  7. Hub, S. I. Springer Nature unveils two new AI tools to protect research integrity (2024).

Cited by

See all 40

Metadata

Authors
Porter, S. J., & McIntosh, L. D.
Affiliations
Digital Science (United Kingdom) (GB)
Corresponding author
Porter, Simon J.
Venue
Scientific Reports
Published
2024-11-28
Volume
14
Issue
1
Pages
29569-29569
ISSN
2045-2322 (ISSN-L)
OpenAlex
W4404830201
Cited by
22
References
34
Citations by year
2026: 8 · 2025: 11 · 2024: 3
FWCI
9.09
Type
journal article
Language
en
Open access
Licence
Locations
  • Scientific Reports (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • arXiv (Cornell University) (submittedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • arXiv (Cornell University) (cc-by)
Topics
  • Social Sciences › Education › Academic integrity and plagiarism
  • Social Sciences › Library and Information Sciences › scientometrics and bibliometrics research
  • Physical Sciences › Physics and Astronomy › Statistical and Nonlinear Physics › Complex Network Analysis Techniques
Keywords
paper mills, authorship fraud, network fingerprinting, co-authorship networks, research integrity

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-05, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-05 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol and qwen3.8-omni-flash).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Identifying fabricated networks within authorship-for-sale enterprises. https://app.cochat.ai/papers/10.1038/s41598-024-71230-8
The paper
Porter, S. J., & McIntosh, L. D. (2024). Identifying fabricated networks within authorship-for-sale enterprises. Scientific Reports. https://doi.org/10.1038/s41598-024-71230-8

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.