Scholarlink

This paper presents a method for automatically creating ontological knowledge bases from Russian text corpora using advanced Natural Language Processing (NLP) models: Bidirectional Encoder Representations from Transformers (BERT) and Generative Pre-trained Transformer (GPT). Existing ontology construction tools such as Text2Onto, OntoGen, and FRED have limitations, especially with Russian texts. Our approach combines BERT’s ability to precisely identify concepts and relationships with GPT’s strength in generating implicit connections. Experimental evaluations on texts demonstrate that our combined model significantly improves recall and F1-measure compared to existing methods. The practical implementation resulted in a comprehensive ontology that clearly represented key concepts and their interactions. The paper also discusses integrating these ontologies with knowledge management platforms such as OntoDog and Stardog, highlighting practical usability. Future research directions include improving quality control mechanisms and adapting the proposed method for multilingual text analysis.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Using BERT and GPT as an Effective Solution for Ontology Development Automation

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Alexander Katyshev,
  • Anton Anikin,
  • Vladislav Smirnov

摘要

This paper presents a method for automatically creating ontological knowledge bases from Russian text corpora using advanced Natural Language Processing (NLP) models: Bidirectional Encoder Representations from Transformers (BERT) and Generative Pre-trained Transformer (GPT). Existing ontology construction tools such as Text2Onto, OntoGen, and FRED have limitations, especially with Russian texts. Our approach combines BERT’s ability to precisely identify concepts and relationships with GPT’s strength in generating implicit connections. Experimental evaluations on texts demonstrate that our combined model significantly improves recall and F1-measure compared to existing methods. The practical implementation resulted in a comprehensive ontology that clearly represented key concepts and their interactions. The paper also discusses integrating these ontologies with knowledge management platforms such as OntoDog and Stardog, highlighting practical usability. Future research directions include improving quality control mechanisms and adapting the proposed method for multilingual text analysis.