Scholarlink

<p>In line with the emerging paradigm of academic publishing known as open peer review, this study investigates the utility and effectiveness of peer review information in predicting the citation impact of publications using an interpretable machine learning approach. Through a literature review, we defined three citation metrics as the dependent variables and 24 features categorized as paper- and review-related as the independent variables. We trained machine learning models to predict citation metrics using data from 11,861 articles published in <i>eLife</i> between 2013 and 2022. The model that integrated both paper- and review-related features predicted citation metrics by 1.00% to 3.74% more accurately than the model that used only paper-related features, and we statistically verified the significance of these differences. We also applied the SHapley Additive exPlanation (SHAP) technique to interpret the influence of each feature on the model’s predictions. The submission-to-acceptance dates and the number of disclosed reviewers emerged as the most crucial features, both showing a negative relationship with citation metrics. These findings contribute to the literature by demonstrating the utility of author response data, clarifying the role of peer review information in shaping the citation impact of publications, and offering new insights into the academic publishing process.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Unveiling the potential of peer review information in predicting citation impact

  • 期刊论文

  • 发表日期：

    2026年3月24日

查看全文
  • Jaemin Chung,
  • Seunghyun Lee,
  • Janghyeok Yoon,
  • Wan Jong Kim

摘要

In line with the emerging paradigm of academic publishing known as open peer review, this study investigates the utility and effectiveness of peer review information in predicting the citation impact of publications using an interpretable machine learning approach. Through a literature review, we defined three citation metrics as the dependent variables and 24 features categorized as paper- and review-related as the independent variables. We trained machine learning models to predict citation metrics using data from 11,861 articles published in eLife between 2013 and 2022. The model that integrated both paper- and review-related features predicted citation metrics by 1.00% to 3.74% more accurately than the model that used only paper-related features, and we statistically verified the significance of these differences. We also applied the SHapley Additive exPlanation (SHAP) technique to interpret the influence of each feature on the model’s predictions. The submission-to-acceptance dates and the number of disclosed reviewers emerged as the most crucial features, both showing a negative relationship with citation metrics. These findings contribute to the literature by demonstrating the utility of author response data, clarifying the role of peer review information in shaping the citation impact of publications, and offering new insights into the academic publishing process.