首页|期刊导航|河北工业科技|基于RAG与领域术语增强的电力通信客服知识库构建方法

基于RAG与领域术语增强的电力通信客服知识库构建方法OA

Construction method for power communication customer service knowledge base based on RAG and domain terminology enhancement

中文摘要英文摘要

为了解决通用大语言模型在电力通信客服领域中存在的事实性错误与领域适应性差问题,提出一种基于检索增强生成(retrieval-augmented generation,RAG)与领域术语增强的电力通信客服知识库自动构建方法.首先,通过无监督新词发现与领域词典构建技术,自动形成电力通信领域术语库,并据此优化检索系统的语义理解能力;其次,设计一种分层检索策略,结合术语关键词与语义向量,自动从标准文档和历史问答中精准检索相关的知识片段;最后,利用检索到的信息作为增强上下文,通过动态指令模板自动引导大语言模型生成准确、规范的问答对,并通过闭环优化机制自动提升生成质量.结果表明:所构建的电力通信客服知识库术语规范率达94.2%,参数准确率达92.8%,标准引用准确率达86.7%;领域术语增强与分层检索策略对性能提升贡献显著,所生成的问答对质量显著优于传统直接生成式方法,有效缓解了模型幻觉问题.所提方法可为电力通信等垂直行业的智能化客服建设提供可靠的技术路径.

In order to address the issues of factual inaccuracies and poor domain adaptability in the field of electric power communication customer service,The power communication customer service sector is highly specialized,characterized by extensive terminology and rapidly evolving knowledge.The direct application of general large language models in this domain often leads to factual inaccuracies and poor adaptability.To address these challenges,this paper proposed an automatic knowledge base construction method for electric power communication customer service based on retrieval-augmented generation(RAG)and domain terminology enhancement was proposed.Firstly,unsupervised new word discovery and domain dictionary construction techniques were employed to automatically build a domain-specific terminology library for power communication,which subsequently optimized the semantic understanding capabilities of the retrieval system.Secondly,a hierarchical retrieval strategy was designed,integrating key terminology with semantic vectors to enable precise and automatic retrieval of relevant knowledge fragments from standard documents and historical question-answer data.Finally,using the retrieved information as enriched context,a dynamic instruction template was developed to automatically guide a large language model in generating accurate and standardized question-answer pairs.A closed-loop optimization mechanism further enhanced the quality of the generated output.Experimental results show that the constructed power communication customer service knowledge base achieves a terminology standardization rate of 94.2%,a parameter accuracy of 92.8%,and standard citation accuracy of 86.7%.Domain terminology enhancement and the hierarchical retrieval strategy significantly contribute to the performance improvements.The generated question-answer pairs substantially outperform those produced by baseline direct generation models in terms of quality,effectively mitigating the problem of model hallucination.The proposed method provides a reliable technical pathway for the development of intelligent customer service systems in vertical industries such as power communication.

许密;李瑾辉;胡皓;符士侃;丁士长

国网江苏省电力有限公司信息通信分公司,江苏 南京 210024国网江苏省电力有限公司信息通信分公司,江苏 南京 210024国网江苏省电力有限公司信息通信分公司,江苏 南京 210024国网江苏省电力有限公司信息通信分公司,江苏 南京 210024国网江苏省电力有限公司信息通信分公司,江苏 南京 210024

信息技术与安全科学

人工智能大语言模型知识库构建电力通信检索增强生成领域术语增强

artificial intelligencelarge language model(LLM)knowledge base constructionpower communicationretrieval-augmented generation(RAG)domain terminology enhancement

《河北工业科技》 2026 (4)

328-339,12

国家电网科技项目(J2024083)

10.7535/hbgykj.2026yx04005

评论