Abstract:

Purposes This study examines the characteristics and drivers of preprint withdrawals to inform governance strategies for preprint platforms and their collaborative development with academic journals. Methods Leveraging a dataset of 19767 retracted preprints from arXiv （from 1991 to 2024）， we employed quantitative and textual analysis to examine withdrawal characteristics across temporal， disciplinary， and collaborative dimensions， and utilized zero-shot classification to automatically identify and categorize the reasons for retraction. Findings Our analysis revealed that solo-authored and small-team preprints were retracted more frequently， mostly within a year of posting， while larger teams demonstrated greater stability. Combinations of disciplines like Computer Science and Statistics showed a higher incidence of retraction. Furthermore， author-initiated retractions were predominantly for errors， in contrast to platform-initiated retractions， which largely concerned academic misconduct. The reasons for author-initiated retractions are closely associated with paper characteristics： papers with multiple authors and interdisciplinary backgrounds are more likely to be retracted due to author disputes， while single-discipline， low-version papers are primarily retracted due to errors； platform-initiated retractions show no such association. Conclusions Preprint retractions are influenced by collaboration scale， disciplinary field， and the initiating entity. This study provides empirical support for building early-warning mechanisms on preprint platforms and optimizing the peer-review process for journals， offering insights for constructing a research integrity framework within the open science ecosystem.

Key words: Preprint, Journal publishing, Academic integrity, Open science, Academic exchange

摘要：

目的 探讨预印本撤稿现象的特征与动因，为预印本平台的治理及其与学术期刊的协同发展提供参考。 方法 基于arXiv平台1991—2024年19767条撤稿数据，运用定量与文本分析方法，从时间、学科及作者合作模式等维度解析撤稿特征，并采用零样本分类技术对撤稿原因进行自动识别与归类。 结果 独作和小团队的预印本更易发生撤稿，且撤稿多集中于公开后1年内；计算机科学和统计学等学科组合撤稿相对高发；大团队合作有助于维持成果的稳定性；作者申请撤稿主要因研究错误，而平台实施的撤稿主要是针对学术不端行为。作者撤稿原因与论文特征密切关联：多作者、跨学科论文易因作者争议撤稿，而单一学科、低版本论文则主要因存在错误撤稿；平台撤稿则无此类关联。 结论 预印本撤稿行为受合作规模、学科领域、撤稿主体及论文特征的影响，本研究为预印本平台风险预警机制构建与期刊审稿流程优化提供依据，对推动开放科学生态下的科研诚信体系建设具有启示意义。

关键词: 预印本, 期刊出版, 学术诚信, 开放科学, 学术交流