Scholarlink

To improve the accuracy and authenticity of a large language model (LLM) in identifying Policy-Attention themes, this study proposes a method for constructing a domain-specific LLM using LoRA fine-tuning technology. The main steps are as follows: constructing a directive dataset for automotive companies’ responses to new energy policies; LoRA fine-tuning of the ChatGLM4-9B model results in ChatGLM-LoRA-NEVPR; evaluating the model through objective, human, and intelligent assessments. Experimental results show that the proposed method improves the accuracy by 12% compared to the baseline model. Additionally, the results from both human and intelligent evaluations further validate the effectiveness of the method across various dimensions. The paper develops ChatGLM-LoRA-NEVPR, and fine-tuning large language models can significantly enhance its accuracy and authenticity in recognizing policy-attention themes within the new energy vehicle sector.

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

A Fine-Tuned LLM Method to Explore Corporate Policy-Attention Themes in the New Energy Vehicle Sector

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Huiru Jia,
  • Hanfei Li,
  • Jiajia Cai,
  • Daqing Yin,
  • Shengyuan Yin,
  • Yue Han,
  • Zhen Zhu

摘要

To improve the accuracy and authenticity of a large language model (LLM) in identifying Policy-Attention themes, this study proposes a method for constructing a domain-specific LLM using LoRA fine-tuning technology. The main steps are as follows: constructing a directive dataset for automotive companies’ responses to new energy policies; LoRA fine-tuning of the ChatGLM4-9B model results in ChatGLM-LoRA-NEVPR; evaluating the model through objective, human, and intelligent assessments. Experimental results show that the proposed method improves the accuracy by 12% compared to the baseline model. Additionally, the results from both human and intelligent evaluations further validate the effectiveness of the method across various dimensions. The paper develops ChatGLM-LoRA-NEVPR, and fine-tuning large language models can significantly enhance its accuracy and authenticity in recognizing policy-attention themes within the new energy vehicle sector.