Ontology-based Knowledge Graph for Industrial Standards with Hierarchical and Conditional Structuring

ACL ARR 2026 January Submission4649 Authors

Keywords: Knowledge Graph, Triplet Extraction, Retrieval-Augmented Generation (RAG), Industrial Standards Documents, Ontology-based Knowledge Graph

Abstract:

Industrial standards documents contain complex conditional statements and table-based rules, making them challenging to interpret using conventional text-based approaches.
Although large language model–based document understanding methods have been actively studied, they often fail to capture hierarchical and conditional document structures, limiting their ability to model complex conditions and table-driven rules.
To address this challenge, we propose an Ontology-based knowledge graph (KG) construction method that integrates ontologies with conditional-structure-based triplet extraction, and evaluate its effectiveness using question answering datasets constructed from multiple industrial standards documents.
Experimental results show statistically significant performance improvements over baseline models, with particularly notable gains on reasoning tasks involving table-based rules and multi-condition reasoning, demonstrating that the proposed Ontology-based KG effectively captures document structure for reliable question answering. Code is available at: https://anonymous.4open.science/r/ontology_based_kg_paper-F5CE

Paper Type: Long

Research Area: Information Extraction and Retrieval

Research Area Keywords: knowledge graphs, knowledge base construction, document-level extraction, document representation

Contribution Types: Model analysis & interpretability, NLP engineering experiment, Data analysis

Languages Studied: English

Submission Number: 4649