Integrity Review Report: Boosting Semi-Supervised Video Captioning via Learning Candidates Adjusters (DOI: 10.1145/3652838)

Academic fraud report · Geng Detector

Summary

This report assesses the academic integrity of an ACM TOMM 2024 paper on semi-supervised video captioning by Xu et al. The verdict is HIGHLY SUSPICIOUS due to severe editorial and typographic errors, rather than evidence of data fabrication. The most critical findings are: (1) a temporal paradox in the publication timeline where the revised date (10 July 2024) post-dates the accepted date (25 February 2024); (2) author affiliations listing Beijing Jiaotong University as being located in 'Beijing, United States', likely a template fill error; and (3) numerous spelling/OCR artifacts in the manuscript text such as 'd e ine', 'irst', 'pesudo labels', 'eectiveness', and 'luency'. Notably, cross-table consistency checks (Tables 1 and 3) showed internally consistent numerical values, and performance fluctuations in Table 5 (e.g., CIDEr dropping from 95.03 to 94.09 at higher training steps) are consistent with genuine experimental noise, suggesting the data is not fabricated. Image reuse could not be assessed. Confidence is high regarding the textual errors but limited regarding deeper misconduct due to lack of pixel-level image access.

Verdict

Highly Suspicious (🟠) — primarily driven by unprofessional editorial errors (timeline paradox, incorrect country affiliations, pervasive spelling mistakes) rather than detected data fabrication. The combination of such basic errors in a high-profile ACM venue publication by senior authors (including IEEE Fellow Qiang Ji) severely undermines confidence in the manuscript's review process.

Key findings

Publication timeline paradox: The paper states "Received 5 August 2023; revised 10 July 2024; accepted 25 February 2024." The revision date (10 July 2024) occurs after the acceptance date (25 February 2024), which is logically impossible in standard peer-review workflows.
Geographic/affiliation errors: Multiple authors at Beijing Jiaotong University are listed with address "Beijing, United States," and one author (LILI WAN) appears with "Beijing, Beijing, United States."
Extensive spelling/OCR artifacts: Numerous text errors including "d e ine" (define), "irst" (first), "pesudo labels" (pseudo), "eectiveness" (effectiveness), and "luency" (fluency).
Cross-table consistency verified: Table 1 LCABM(100%) MSR-VTT metrics (B@4=46.35, R=63.01, M=29.06, C=52.8) match Table 3 beam3 results exactly.
Plausible performance noise: Table 5 (MSVD) shows CIDEr dropping from 95.03 (T=2 E-step) to 94.09 (T=3 E-step), consistent with realistic overfitting/instability rather than fabricated monotonic gains.
Last-digit distribution: Metric values such as 52.8, 94.3, 61.39, and 77.04 show natural variability without obvious integer-preference patterns.

Evidence highlights

Timeline contradiction (verbatim): "Received 5 August 2023; revised 10 July 2024; accepted 25 February 2024."
Affiliation error (verbatim examples):
ZHENJIANG MIAO: "...Beijing Jiaotong University..., Beijing, United States"
YIGANG CEN: "...Beijing Jiaotong University..., Beijing, United States"
LILI WAN: "...Beijing Jiaotong University..., Beijing, Beijing, United States"
Text artifacts (verbatim): "d e ine", "irst to explore", "pesudo labels", "eectiveness", "luency"
Numerical consistency check: Table 1 LCABM(100%) MSR-VTT {B@4=46.35, R=63.01, M=29.06, C=52.8} ≡ Table 3 beam3.
Realistic performance dip: Table 5 MSVD CIDEr at T=2 E-step = 95.03; T=3 E-step = 94.09.

Notes

DOI: 10.1145/3652838
Venue/year: ACM Trans. Multimedia Comput. Commun. Appl. (ACM TOMM), 2024.
Image analysis limitation: Pixel-level inspection of Figures 1–5 was not possible from the provided text-based source; image reuse or splicing cannot be ruled in or out.
Uncertainty: The timeline and country errors may stem from careless production editing rather than author intent; however, their presence in a final published article points to a failed editorial/review process. No statistical fraud indicators were detected.
Recommended action: Issue an Erratum to correct affiliation country and publication dates; this report does not constitute formal misconduct determination.