多模型协同驱动的学术论文智能编校工作流构建与效能评价
Construction and efficiency evaluation of an intelligent editing and proofreading workflow for academic papers collaboratively driven by multiple large language models
摘要
本研究旨在构建一个由多模型协同驱动的智能编校工作流,以提升编校质量、效率和自动化水平.基于5类编校差错案例库,对10种具备长文本处理能力的大语言模型开展性能测评,依托Dify平台构建多模型协同编校工作流(MLC-EPW),并选取2篇具有代表性的学术论文作为测试对象进行实证对比分析.MLC-EPW的编校质量最优,总体差错正确识别率和修正率均最高,且差错错误识别率较低;MLC-EPW的编校效能最高,处理单篇论文仅需约12 min,成本11~13元/篇.本研究构建的MLC-EPW有效整合了不同大语言模型的优势,在编校质量、效率与成本方面均优于单一模型方案,为学术期刊实现智能化、自动化编校流程提供了可行的技术路径与实践参考.
This study endeavors to devise a multi-LLM(Large Language Model)collaboratively powered intelligent editing and proofreading workflow,with the objective of elevating the quality,efficiency,and automation of the editing and proofreading procedures.Utilizing a case library encompassing five categories of editing and proofreading errors,we conducted performance evaluations on ten large language models renowned for their long-text processing capabilities.Leveraging the Dify platform,we constructed a multi-LLM collaborative editing and proofreading workflow(MLC-EPW)and selected two representative academic papers as test cases for an empirical comparative analysis.The MLC-EPW exhibited exceptional editing and proofreading performance,achieving the highest overall accuracy rates in error detection and correction while maintaining a comparatively low false-positive rate.Regarding efficiency,the MLC-EPW processed individual papers in around 12 minutes,with costs fluctuating between 11 and 13 CNY per paper.The MLC-EPW developed in this study effectively harnesses the combined strengths of various LLM,surpassing single-model methods in terms of editing and proofreading quality,efficiency,and cost-efficiency.This offers a viable technical pathway and practical reference for academic journals to implement intelligent and automated editing processes.
年,卷(期):
2026, 38(4)
机标分类号:
G237.5
TP391
F272.92
在线出版日期:
2026年9月24日
基金项目:
2025ZBCH-JYZD05:中华出版促进会研究课题CESSP-CJWK-2025005:中国科学技术期刊编辑学会长江文库计划编辑学研究项目,