• Home
  • Search
  • Classifying Scientific Peer Reviews: Distinguishing Authentic, Generic, and AI-Generated Feedback
  • https://doi.org/10.1109/fmlds67896.2025.00068Copy DOI Icon

Classifying Scientific Peer Reviews: Distinguishing Authentic, Generic, and AI-Generated Feedback

  • Nov 2, 2025
  • Muhammad Aashir Aftab +4 more
Show More
  • Abstract
  • Literature Map
  • References
  • Similar Papers
Abstract

The quality, and more importantly, the integrity of academic research heavily rely on the peer review process. However, the increasing sophistication and accessibility of Large Language Models (LLMs) pose a potential threat, as they can be utilized to generate generic and ineffective peer reviews, thereby undermining the rigor of scholarly publications. "Pal-reviews," providing generic comments with no critical feedback, are another pressing subject in scientific publications. This research investigates the feasibility of automatically auditing and categorizing peer reviews using fine-tuned transformer-based language models. We explored five prominent architectures, i.e., BERT, XLNet, DistilBERT, RoBERTa and OpenAI GPT-4o-nano, while training them on a very specifically curated dataset of authentic, AI-generated, and generic reviews. We generated the authentic reviews from journals that publish open reviews with articles. ChatGPT was used to generate AI-generated reviews, while sentence concatenation was used to generate generic reviews. Our methodology involved combining paper titles, abstracts, and review text with special tokens, employing robust text encoding techniques, and utilizing a rigorous training routine incorporating cross-validation, focal loss for class imbalance, and early stopping. The models were evaluated using accuracy, F1-score, precision, and recall. Our findings indicate the promising potential of this approach, with GPT-4o-nano substantially surpassing all other models, achieving a held-out test accuracy of 80.7% and macro-F1 of 80.6%. RoBERTa, DistilBERT and XLNet followed with moderate performance, while BERT lagged behind. To encourage reproducibility, the fine-tuned the BERT based models have been published on the Hugging Face Model Hub with open access.

Similar Papers
  • Front Matter
  • Citations9

Peer Review: The Best of the Blemished?

  • Mar 29, 2007
  • The American Journal of Medicine
  • Joseph S Alpert
  • Research Article
  • Citations5

Moving toward a more objective peer review process

  • Jan 01, 2014
  • Nursing Management
  • Kristin Ray +1
  • Research Article
  • Citations6

Peer review under the spotlight in the UK

  • Mar 13, 2012
  • Journal of Advanced Nursing
  • Roger Watson
  • Dataset
  • Citations6

Reproducibility: does it really matter?

  • Jun 09, 2016
  • The Winnower
  • Jaime A Teixeira Da Silva
  • Research Article

Networks of Editors: Useful for Researchers

  • Jun 01, 2016
  • Nurse Author & Editor
  • Helena Leino‐Kilpi
  • Research Article
  • Citations3

Peer Review in Transplantation.

  • Sep 01, 2015
  • Transplantation
  • James A Hutchinson
  • Research Article

Peer Review and the PJOHNS: Principles, Problems, and Promise

  • Jul 12, 2018
  • Philippine Journal of Otolaryngology-Head and Neck Surgery
  • José Florencio F Lapeña
  • Research Article
  • Citations2

Naming peer reviewers in JRSM

  • Oct 01, 2009
  • Journal of the Royal Society of Medicine
  • Kamran Abbasi
  • Research Article

Editorial Policy

  • Nov 19, 2012
  • Liceo Journal of Higher Education Research
  • R Deepa
  • Research Article
  • Citations3

How to be a good reviewer: A step-by-step guide for approaching peer review of a scientific manuscript.

  • Jun 01, 2024
  • Laryngoscope investigative otolaryngology
  • Ahmad R Sedaghat +5
  • Front Matter
  • Citations1

ANZJOG in 2016: Passing the baton.

  • Jan 27, 2016
  • The Australian & New Zealand journal of obstetrics & gynaecology
  • Jan E Dickinson
  • PDF
  • Research Article
  • Citations4

Peer review reduces spin in PCORI research reports

  • Dec 01, 2021
  • Research Integrity and Peer Review
  • Evan Mayo-Wilson +5
  • Research Article
  • Citations2

Examining peer review network dynamics in higher education visual communication courses using ERGM

  • Sep 30, 2024
  • Computers and Education Open
  • Ly Dinh +2
  • Research Article

Deciding to post a manuscript preprint

  • Dec 01, 2020
  • Nurse Author & Editor
  • Jessica Castner +3
  • Research Article
  • Citations16

The ethics of peer review in bioethics

  • Oct 16, 2013
  • Journal of Medical Ethics
  • David Wendler +1
Cactus Communications logo

Copyright 2026 Cactus Communications. All rights reserved.