使用大语言模型辅助论文同行评议的实验  

An Experiment Using a Large Language Model to Assist Paper Peer Review

作　　者：王凌峰[1] 任雅梅 WANG Lingfeng;REN Yamei

机构地区：[1]桂林电子科技大学商学院,广西桂林541004

出　　处：《图书情报导刊》2025年第1期48-54,共7页Journal of Library and Information Science

摘　　要：讨论大语言模型在论文同行评议中的应用,提出大语言模型辅助同行评议的通用步骤。以发表年度修正论文被引次数的预测正确率作为评价指标,使用国产月之暗面大语言模型对国内预印本主题研究论文进行机器评议实验。实验设定大语言模型对论文评分的不同门槛值,对比各门槛值下发表年度修正论文被引次数的预测正确率变化,结果显示随着大语言模型论文评分门槛值的提高,预测正确率明显增加,表明大语言模型可用于辅助论文同行评议。This study explores the application of large language models(LLMs)in peer review of academic papers,and proposes a general framework for LLM-assisted peer review.Using prediction accuracy of annual citation counts as an evaluation metric,this study conducts an experimental by using Chinese-developed Moonshot AI large language model.The experiment sets different threshold values for LLM scoring on papers and compares the changes in prediction accuracy of annual citation counts on different threshold levels.Results demonstrate that as LLM's scoring threshold increases,prediction accuracy significantly improves,indicating that large language models can effectively assist in peer review process of academic papers.

关 键 词：同行评议 大语言模型 发表年度 被引次数 实验 

分 类 号：G237.5[文化科学]