| 摘 要: 针对离散生产客诉服务中技术文档非结构化、业务数据割裂以及大模型在该领域易产生“幻觉”的问题，设计了基于图检索增强生成的智能问答系统。提出了面向质量追溯系统的生产缺陷知识图谱与领域知识库构建方法，并融合Leiden社区检测算法与混合检索策略，增强了问答系统的准确性与可解释性。实验采用某轴承企业的售后历史数据验证，在RAGAs评估框架中较通用GraphRAG方法综合得分提升了8.60分，验证了其在专业领域知识抽取、组织与应用方面的可行性与实用性。 |
| 关键词: 离散生产 领域知识库 图检索增强 大模型 |
|
中图分类号: TH186;TP18
文献标识码:
|
|
| Graph Retrieval-Augmented Generation for Intelligent Customer Complaint Question-Answering System in Discrete Manufacturing |
|
CAI Jiajun, BAO Min
|
Zhejiang Sci-Tech University
|
| Abstract: Addressing the challenges of unstructured technical documents, fragmented business data, and the susceptibility of large language models to 'hallucinations' in discrete manufacturing customer-complaint services, this paper designs an intelligent question-answering system based on graph retrieval-augmented generation. It proposes construction methods for a production-defect knowledge graph and a domain knowledge base tailored to quality-traceability systems, and integrates the Leiden community detection algorithm with a hybrid retrieval strategy to enhance the system’s accuracy and interpretability. Experiments conducted on after-sales historical data from a bearing manufacturer demonstrate an 8.60-point increase in the overall RAGAs evaluation score compared with the general GraphRAG method, validating the system’s feasibility and practicality in domain-specific knowledge extraction, organization, and application. |
| Keywords: Discrete Manufacturing Domain Knowledge Base Graph Retrieval-Augmented Generation Large Language Model |
