大模型赋能的供应商注册管理本体构建方法研究

Research on Large-Model Empowered Ontology Construction for Supplier Registration Management

摘要: 针对人工智能与数字化转型背景下供应商管理领域存在的知识体系缺失及人工依赖度高等问题，提出基于大模型与基本形式本体(BFO, Basic Formal Ontology)相结合的供应商注册管理本体构建方法。选用通用开源大模型作为核心语义处理底座，配套检索增强RAG模块完成领域文档定向知识抽取，通过将大模型的语义理解与知识抽取能力嵌入七步法本体构建流程，构建包含术语智能抽取、层级动态映射、属性约束及推理验证等关键环节的闭环体系。设计分层级提示词工程实现多环节精准语义解析，配套人机协同修正模型事实偏差、关系混淆等错误输出，界定模型自动化处理与领域专家校验的分工边界，形成持续迭代闭环工作流。研究采用推理机对本体进行验证，成功实现对供应商股权控制识别等复杂场景的逻辑推理功能，有效验证了构建方法与本体框架的可行性。该方法形成标准化知识框架，突破传统经验管理模式，为供应商注册管理领域实现从经验驱动向数据驱动与知识驱动的范式转型提供了可复用的技术路径。

Abstract: Against the backdrop of artificial intelligence and digital transformation, supplier management is confronted with issues including the lack of a sound knowledge system and high reliance on manual work. To address these problems, this paper proposes an ontology construction method for supplier registration management that integrates large language models (LLMs) with the Basic Formal Ontology (BFO). A general open-source large language model is adopted as the core semantic processing foundation, supplemented by a Retrieval-Augmented Generation (RAG) module to conduct targeted knowledge extraction from domain documents. By embedding the semantic understanding and knowledge extraction capabilities of LLMs into the seven-step ontology construction process, a closed-loop system covering key links such as intelligent term extraction, dynamic hierarchical mapping, attribute constraint definition and reasoning verification is established. Hierarchical prompt engineering is designed to realize accurate semantic parsing across multiple stages. A hu-man-machine collaborative correction mechanism is deployed to rectify model output errors such as factual deviations and confused relational definitions, define the division of responsibilities between automated model processing and domain expert verification, and form a continuously iterative closed-loop workflow. An inference engine is utilized for ontology validation in this research, which successfully enables logical reasoning for complex scenarios such as the identification of equity control among suppliers, thereby effectively verifying the feasibility of the proposed construction method and ontology framework. This method establishes a standardized knowledge framework and breaks through the traditional experience-based management paradigm. It provides a reusable technical path for supplier registration management to achieve paradigm transition from experience-driven governance to data-driven and knowledge-driven governance.