Scholarlink

<p>Knowledge graph technology has significantly advanced in recent years in solving a variety of industrial issues, including intelligent decision-making, defect diagnostics, root cause analysis, and quality traceability. However, knowledge graphs frequently contain a sizable number of missing items as a result of inadequate industrial informatization. In industrial knowledge graphs, completing entities is uncommon. In this study, we present an NFET-GAN approach for completing the entity completion problem in industrial knowledge graphs using a zero-shot learning framework. Here, missing entities’ semantic characteristics are inferred from their textual descriptions. The one-hop range of neighborhood features taken into account in this study actually enhances intra-class similarity while decreasing inter-class similarity. The suggested technique surpasses previous approaches, according to experimental data, which provide an average improvement in Mean Reciprocal Rank of 61.4 percent on real assembly domain knowledge. This method specifically improves performance in knowledge graphs from the universal encyclopedia, demonstrating the universality of the methodology.</p>

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

Entity completion for industrial knowledge graph based on zero-shot learning

  • 期刊论文

  • 发表日期：

    2025年12月9日

查看全文
  • Yin Cai,
  • Zhijun Fang,
  • Yuanyuan Li,
  • Jiacheng Xu,
  • Anjie Wang,
  • Zheyi Cheng

摘要

Knowledge graph technology has significantly advanced in recent years in solving a variety of industrial issues, including intelligent decision-making, defect diagnostics, root cause analysis, and quality traceability. However, knowledge graphs frequently contain a sizable number of missing items as a result of inadequate industrial informatization. In industrial knowledge graphs, completing entities is uncommon. In this study, we present an NFET-GAN approach for completing the entity completion problem in industrial knowledge graphs using a zero-shot learning framework. Here, missing entities’ semantic characteristics are inferred from their textual descriptions. The one-hop range of neighborhood features taken into account in this study actually enhances intra-class similarity while decreasing inter-class similarity. The suggested technique surpasses previous approaches, according to experimental data, which provide an average improvement in Mean Reciprocal Rank of 61.4 percent on real assembly domain knowledge. This method specifically improves performance in knowledge graphs from the universal encyclopedia, demonstrating the universality of the methodology.