面向标准数字化的语义知识库自动构建技术研究OA

To address the core challenges of fragmented standard documents,lack of semantic correlation,and poor machine readability,and in response to the strategic requirements for the digital transformation of standardization outlined in the National Standardization Development Outline,this paper proposes a semi-automatic construction method for a standard semantic knowledge base that integrates domain ontology and deep learning technologies.First,through systematic domain analysis and formal modeling,a quintuple conceptual model centered on"standardized object-stylistic norm-indicator item-indicator value-qualifier class"is constructed,providing a unified framework for machine-readable knowledge expression.Second,a two-phase technical architecture is designed and implemented:In the first phase,a joint extraction model based on domain-adaptive pre-training and rule guidance is developed,enabling the accurate identification and structuring of key knowledge triples from unstructured standard texts;In the second phase,graph neural networks are introduced for knowledge representation learning,automatically mining and complementing potential deep semantic correlations through link prediction tasks,thereby optimizing the structural integrity and semantic richness of the knowledge graph.Finally,an empirical study is conducted with safety and environmental standards in the agricultural and food domain as the dataset.Experimental results show that the proposed method achieves an F1-score of 89.7％in the knowledge element extraction task and can effectively construct a well-structured knowledge network rich in semantic associations.The core contributions of this research are:systematically proposing for the first time a large-scale automated computational method for semantic correlation oriented to standard content;establishing a universal specification for standard knowledge representation;significantly enhancing the interoperability and reuse value of digitalization outcomes across different standard domains;and laying a high-quality,structured data foundation for downstream advanced applications such as intelligent Q&A and compliance review.