A 501(c)(3) non-profit organization info@cheairesearch.com
Applied AI research for public benefit

Three articles retracted after fabricated references were documented

VNU Journal of Science: Education Research has retracted three articles following reports we submitted documenting references that could not be verified against any independent record. This note sets out what we checked, how we checked it, and what the journal did.

The three retracted articles

All three now carry the RETRACTED label in the journal record. The links below go to the journal pages.

Retracted

Learning-Oriented Assessment in Higher Education: From Theory to Practice in Developing Vietnamese Students’ Competencies

Vol 41, No 3 · Original article · Published 24 September 2025

js.vnu.edu.vn/ER/article/view/5335
Retracted

Approaches to Applying Deep Learning for Early Prediction of Learners’ Academic Outcomes

Vol 41, No 4 · Review article · Published 12 November 2025

js.vnu.edu.vn/ER/article/view/4967
Retracted

Challenges Faced by Non-major Students in Learning English to Achieve the Outcomes at a Vietnam Public University

Vol 41, No 4 · Original article · Published 3 July 2025

js.vnu.edu.vn/ER/article/view/4978

How the references were checked

The work was carried out under VeritasAudit, our verification project. A reference is only reported when it survives every stage below, because a citation can fail an automated lookup for ordinary reasons such as a typo, an unusual transliteration, or a journal that is simply not indexed.

01

Parse the reference list

Each article was retrieved from the journal and its reference list parsed into structured fields covering authors, title, journal, volume, issue, pages, year, and DOI where present.

02

Match against bibliographic databases

Every parsed reference was queried against CrossRef and OpenAlex. A match requires agreement on title and author, not merely a similar phrase.

03

Rescue pass before any report

References with no database match were re-checked against Google Scholar. This stage matters. In the wider batch it moved 108 references out of the suspected category, and reporting without it would have produced false accusations.

04

Manual confirmation at the source

Remaining items were checked by hand at the named journal. In several cases the cited volume, issue, and page range exist but belong to an entirely different article, which is the clearest evidence that a reference was constructed rather than consulted.

05

Report to the journal

Findings were sent to the editorial board with the evidence attached, item by item, so each claim could be checked independently.

A deliberate limit on scope. Of the 146 articles collected from this journal, 96 were in Vietnamese and were excluded from automated screening. Their references are not reliably indexed in CrossRef or OpenAlex, so an automated pass would report absence as fabrication. Only the 50 English-language articles were screened.

What the screening found

Across the 50 English-language articles, 1,504 references were examined. The figures below are the final verdicts after the rescue pass, not the raw output of the first automated run.

1,504
References examined
50
Articles screened
436
Could not be verified
108
Cleared at rescue stage

Two of the three retracted articles fell well outside the normal range. In one, 15 of the 23 references taken to the final check stage could not be traced to any independent record. In another, the figure was 9 of 11. A single unverifiable citation invites a correction. A reference list where most entries cannot be traced points to a different problem.

The pattern we look for

Fabricated references are rarely obvious in isolation. They tend to name real authors who work in the field, real journals, and plausible titles. Three signatures recur.

  • Occupied coordinates. The volume, issue, and page range are real but already belong to a different article by different authors on a different subject.
  • Circular provenance. A search for the cited title returns only the article doing the citing, with no independent record of the work anywhere.
  • Plausible but absent. The authors are real researchers in the field and the journal exists, yet the specific article named does not appear in that journal at all.

A citation is a claim that can be checked

A reference list is a promise that the work rests on something. When entries are invented, a reader who tries to follow the argument to its source finds nothing there. Students are taught from these articles. Other researchers build on them. Committees weigh them in promotion decisions.

The problem is also growing more common, because language models produce references that look correct and are not. That is the reason VeritasAudit exists as a tool rather than a one-time exercise. Checking a citation should be fast enough that people actually do it.

We want to be clear about what a retraction is and is not. It corrects the scientific record. It is not a judgment we are in a position to make about any individual, and we have deliberately kept this newsletter focused on the articles and the evidence rather than on the people named in them.

Ongoing work

Screening continues across a wider set of journals. Where we find articles that meet the same evidentiary bar, we report them to the publishing journal first and give the editorial board the opportunity to review before anything is said publicly. That order matters to us. The purpose is a corrected record, not an accusation.

If you edit a journal and would like your reference lists screened, or if you believe an article of yours has been reported in error, please write to us. We will share the underlying evidence for any finding on request.

Questions about this work or the evidence behind it

info@cheairesearch.com