摘要：

【目的】探索大语言模型（Large Language Model，LLM）技术在开放同行评议中的应用潜力，结合评审专家意见和作者回复数据，自动生成元评审（Meta-Review），辅助编辑快速归纳评审要点，助推开放同行评议过程向智能化方向发展。【方法】 基于LLM，利用少样本对比学习、自我反思等提示学习方法，识别评审专家和作者回复的论点和“疑问-回复”论点对，采用分步引导式方法指导LLM自动生成元评审。【结果】 通过合理的提示学习策略，LLM能够有效识别同行评议中的论点以及“疑问-回复”论点对，生成具有论证性结构的元评审内容，显著提升元评审生成的质量。【结论】 基于LLM的元评审归纳和生成可为优化开放同行评议流程和提高评审效率提供有效途径。

Abstract:

[Purposes] This study aims to explore the potential application of large language model (LLM) technology in open peer review. By combining the opinions of review experts and the response data of authors, a meta-review is automatically generated to assist editors in quickly summarizing the key points of review and promote the development of the open peer review process in an intelligent direction. [Methods] Based on the LLM, we utilized prompt learning methods such as few-shot contrastive learning and self-reflection to identify the arguments and "problem-reply" argument pairs between review experts and authors, and used a step-by-step guidance method to guide the model to automatically generate meta-review. [Findings] Through a reasonable prompt learning strategy, the LLM can effectively recognize the arguments and "problem-reply" argument pairs and generate meta-review content with an argumentative structure, which remarkably enhances the quality of meta-review generation. [Conclusions] Meta-review induction and generation based on LLM can offer an effective approach for optimizing the open peer review process and increasing the review efficiency.

中图分类号: 

• 大语言模型

• 开放同行评议

• 元评审

• 提示学习方法

• Large language model

• Open peer review

• Meta-review

• Prompt learning