大语言模型领域意图的精准性增强方法  

Intention recognition accuracy enhancing method for large language model

作　　者：任元凯 谢振平 Ren Yuankai;Xie Zhenping(School of Artificial Intelligence&Computer Science,Jiangnan University,Wuxi Jiangsu 214122,China;Jiangsu Key University Laboratory of Software&Media Technology under Human-Computer Cooperation,Jiangnan University,Wuxi Jiangsu 214122,China)

机构地区：[1]江南大学人工智能与计算机学院,江苏无锡214122 [2]江南大学人机融合软件与媒体技术江苏省高校重点实验室,江苏无锡214122

出　　处：《计算机应用研究》2024年第10期2893-2899,共7页Application Research of Computers

基　　金：国家自然科学基金资助项目(62272201)。

摘　　要：目前通用大语言模型(如GPT)在专业领域问答应用中存在不稳定性和不真实性。针对这一现象,提出了一种在通用大语言模型上耦合领域知识的意图识别精准性增强方法(EIRDK),其中引入了三个具体策略:a)通过领域知识库对GPT输出结果进行打分过滤;b)训练领域知识词向量模型优化提示语句规范性;c)利用GPT的反馈结果提升领域词向量模型和GPT模型的一致性。实验分析显示,相比于标准的GPT模型,新方法在私有数据集上可以提升25%的意图理解准确性,在CMID数据集上可以提升12%的意图理解准确性。实验结果证明了EIRDK方法的有效性。Large language models(such as GPT)exhibit instability and inauthenticity in professional domain Q&A applications.To address this issue,this paper proposed a method to enhance intent recognition by domain knowledge(EIRDK)for large language models.The method involved three specific strategies:a)scoring and filtering the GPT output using a domain knowledge base,b)training the domain knowledge word vector mode to optimize prompt,c)utilizing feedback from GPT to improve the coherence between the domain word vector model and the GPT model.Experimental analysis demonstrates that,compared to the standard GPT model,the new method achieves a 25%improvement in intent understanding accuracy on the private dataset and a 12%increase on the CMID dataset.The results validate the effectiveness of the EIRDK method.

关 键 词：大语言模型知识问答 意图精准性增强 领域知识集成 GPT反馈学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]