Scholarlink

<p>This paper presents a novel framework integrating large-scale multimodal pre-trained models with knowledge graph technologies to advance ceramic design innovation. We propose a comprehensive methodology for constructing domain-specific knowledge graphs that capture the multifaceted nature of ceramic design knowledge across textual, visual, and three-dimensional modalities. The framework employs specialized entity and relation extraction techniques, semi-supervised learning mechanisms, and quality assessment metrics to ensure knowledge graph completeness and accuracy. Building upon this foundation, we develop a cross-domain innovative design reasoning mechanism using graph neural networks with customized message passing and attention mechanisms. Our approach facilitates knowledge transfer between ceramic design and adjacent fields through domain adaptation techniques and multimodal fusion methods. Experimental results demonstrate significant improvements in both knowledge representation quality and innovative design generation, with cross-domain knowledge transfer achieving up to 47% innovation improvement. The proposed framework provides a new paradigm for computational design support that balances domain expertise with creative exploration while respecting material constraints.</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

Large-scale multimodal pre-trained model driven ceramic design knowledge graph construction and cross-domain innovative design reasoning mechanism

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月2日

查看全文
  • Guoxu Zang

摘要

This paper presents a novel framework integrating large-scale multimodal pre-trained models with knowledge graph technologies to advance ceramic design innovation. We propose a comprehensive methodology for constructing domain-specific knowledge graphs that capture the multifaceted nature of ceramic design knowledge across textual, visual, and three-dimensional modalities. The framework employs specialized entity and relation extraction techniques, semi-supervised learning mechanisms, and quality assessment metrics to ensure knowledge graph completeness and accuracy. Building upon this foundation, we develop a cross-domain innovative design reasoning mechanism using graph neural networks with customized message passing and attention mechanisms. Our approach facilitates knowledge transfer between ceramic design and adjacent fields through domain adaptation techniques and multimodal fusion methods. Experimental results demonstrate significant improvements in both knowledge representation quality and innovative design generation, with cross-domain knowledge transfer achieving up to 47% innovation improvement. The proposed framework provides a new paradigm for computational design support that balances domain expertise with creative exploration while respecting material constraints.