Abstract:

[Purposes] This study aims to analyze the trend and formation of retracted papers for fake peer review (FPR) in China, and explore preventive measures. [Methods] Research papers (issued from 2017 to 2021) by Chinese writers on the Retraction Watch website were taken to collect the papers retracted for FPR before August 31, 2022. The global proportion, annual trend, and merged reasons for the retractions in China were analyzed using GraphPad Prism 7.0 software to infer the retraction formation process, in order to propose preventive measures. [Findings] From 2017 to 2021, a total of 635 research papers by Chinese authors were retracted due to FPR, accounting for 52.0% of global research papers retracted due to FPR during the same period. The number of retracted research papers in 2020 and 2021 (95 and 425 respectively) increased significantly compared to that during 2017—2019 (32–43). The main merged reasons for retraction include rogue editors (65.2%), randomly generated content (57.3%), and paper mills (16.5%). Statistical analysis suggests a significant correlation between rogue editors and randomly generated contents (r=0.998, P< 0.001), and both are correlated with the increasing trend of FPR retraction in China (r=0.991, P< 0.001; r> 0.999, P< 0.001). [Conclusions] Machine-written works manipulated by rogue editors for mass publishing are the primary formation approach to FPR retraction in China, leading to escalated retraction in recent years. The products from paper mills serve as a secondary approach to form FPR retraction. To prevent such retraction from the source and path elements, standardizing the writing of artificial intelligence machines, curbing paper mills, and issuing special warnings for journals with management loopholes are feasible measures.

Key words: Retraction due to fake peer review, Rogue editor, Randomly generated content, Academic misconduct, Management strategy

摘要：

【目的】 分析我国同行评议造假撤稿的趋势及形成，并探讨预防对策。【方法】 以撤稿观察网站公布的我国作者2017—2021年发表的研究型论文为研究对象，收集2022年8月31日前因同行评议造假而被撤稿的论文。结合我国此类撤稿的全球表现、趋势和合并的其他撤稿原因，采用GraphPad Prism 7.0分析、推测撤稿形成路径。【结果】 我国同行评议造假撤稿共635篇，占全球同期此类撤稿总量的52.0%。2020年、2021年刊出的因同行评议造假而被撤稿的研究型论文数量（95、425篇）较2017—2019年（32~43篇）明显上升。合并主要的其他撤稿原因依次为流氓编辑（65.2%）、随机生成内容（57.3%）和论文工厂（16.5%）。统计分析提示，流氓编辑与随机生成内容显著相关（r=0.998，P<0.001），二者与我国同行评议造假撤稿数上升趋势相关（r=0.991，P<0.001；r>0.999，P<0.001）。【结论】 机器写作的作品经由流氓编辑操纵而批量发表，是我国同行评议造假撤稿形成的主要路径，导致近年此类撤稿数量上升；论文工厂提供的稿源是撤稿形成的次要路径。规范人工智能机器写作、遏制论文工厂及针对管理存在漏洞的期刊发布专项预警，是我国从源头、路径节点预防此类撤稿的可行对策。

关键词: 同行评议造假撤稿, 流氓编辑, 随机生成内容, 学术不端, 管理策略