基于BERT模型与RLHF的大语言模型协同校对方法研究

打开文本图片集

中图分类号：TP391.1；TP183 文献标识码：A 文章编号：2096-4706（2025）11-0038-06

Research on Collaborative Proofreading Method of Large Language Model Based on BERT Model and RLHF

WU Bian1, YANG Zhengtan²,LI Xiang (1.StateGrid Hubei Electric PowerCo.,Ltd.,Wuhan 430048,China; 2.Wuhan Optics Valley Information Technology Co.,Ltd.,Wuhan 430206, China)

Abstract: The auracy of document proofreading has always faces challenges at the level of complex logic.In order toaleviatethepresureonwritersandfront-linestaff,thisstudyproposesaproofreadingmethodbasedonmulti-model collaboration.Theword-by-wordlabelisgeneratedbyfine-tuningBERTmodel,andtheLargeLanguageModelisfine-tuning usingLoRA tocompensatefordeficienciesindeeperrorunderstanding.ThePPOalgorithm isused tooptimizethedecisionmaking processof te model to met the needsof different scenarios.The multi-modeloutputresultsare integrated through XGBoost toavoidundereporting and misreporting.The experimentalresultsshow thatthis methodhassignifcant advantages in improving the quality and accuracy of document proofreading.

Keywords: document proofreading; BERT;LLM; PPO; XGBoost

0 引言

公文作为党政机关、企事业单位乃至学术机构日常工作中的重要工具，承担着信息传递、决策指导和政策执行的关键任务[1。（剩余9708字）