Academic Integrity Review: "Semi-Supervised Federated Learning via Dual Contrastive Learning and Soft Labeling for Intelligent Fault Diagnosis" (IEEE IoT Journal, 2025)

Academic fraud report · Geng Detector

Summary

Highly suspicious paper published in IEEE Internet of Things Journal (DOI: 10.1109/JIOT.2025.3586718). Multiple textual inconsistencies strongly suggest copy-paste writing and inadequate proofreading. The abstract claims two publicly available datasets while the body cites three; dataset descriptions contain circular self-reference ("same as the PU dataset" when introducing the PU dataset). The private dataset is described as collected from "motors" in the abstract but "pumps" in the experimental section and Figure 2. A single reference [27] is used to describe two unrelated methods in adjacent sentences. A baseline named "FedProx-UDA" is defined using "FedAvg" instead of FedProx. Most strikingly, the paper was submitted on 9 June 2025 and accepted on 30 June 2025 — only 21 days — which is implausible for a top IEEE journal. No image manipulation or code fabrication could be assessed due to lack of access. Verdict: highly suspicious, multiple independent anomalies warrant formal editorial review.

Verdict

🟠 Highly suspicious. The paper shows several independent textual and methodological inconsistencies indicative of rushed authorship, possible LLM-assisted drafting without proper review, or superficial assembly of content. The 21-day acceptance timeline for a top-tier IEEE journal is a major red flag.

Key findings

Inconsistent dataset count: Abstract states "two publicly available datasets"; introduction and experiments use "three openly accessible datasets."
Circular/copy-paste dataset descriptions: The PU dataset description references itself ("same as the PU dataset"); MFPT and CWRU sections use boilerplate "same as ..." references.
Conflicting experimental objects: Abstract says private data was "collected on motors from the factory"; Section IV.A.1 and Figure 2 describe "Reflux pump," "Feeding pump," and "Circulating pump" at a "partner chemical factory" — motors and pumps are distinct mechanical entities with different vibration signatures.
Reused single reference for two unrelated methods: In the contrastive-learning related work, reference [27] is variously cited as both "a contrastive image generation network with multimodal data" and "an online SSL contrastive graph generative network."
Baseline definition error: "FedProx-UDA" is explained as a combination of "FedAvg with UDA" instead of FedProx.
Implausibly fast turnaround: Submitted 9 June 2025; accepted 30 June 2025 — only 21 days, inconsistent with typical IEEE IoT Journal review timelines.

Evidence highlights

DOI: 10.1109/JIOT.2025.3586718
Submission date: 9 June 2025; Acceptance date: 30 June 2025 (21 days)
Abstract: "two publicly available datasets"
Section IV.A.1 introduces PU dataset with self-reference "same as the PU dataset"
Figure 2 caption: "Machines in our partner chemical factory. (a) Reflux pump. (b) Feeding pump. (c) Circulating pump."
Mid-text within 2 sentences, reference [27] is attached to two unrelated methodological claims
Baselines list: "FedProx-UDA: Naive combinations of FedAvg with UDA"

Notes

No figure manipulation or code duplication could be verified because raw images and source code were not accessible for this review.
All findings are textual and metadata-based; they are reproducible from the published PDF.
The 21-day acceptance interval could theoretically be explained by a fast-track arrangement or editorial error, but warrants direct verification by the IEEE IoT Journal editorial office.
Risk of false positives: textual self-references and minor naming errors can occasionally occur in legitimate work, but the combination of five independent anomalies (including a contradictory experimental object) goes beyond ordinary copy-editing oversights.
Recommended actions: contact authors for clarification on the motor-vs-pump discrepancy and dataset count; file a PubPeer comment; request the editorial office to confirm the review record.