Abstract: To increase the efficiency of design knowledge acquisition in the process of conceptual product design,and solve the problems of fragmentation,limited reasoning and insufficient visualisation of data,the design knowledge extraction and knowledge graph construction approach for conceptual product design was proposed based on the joint learning.The design knowledge graph framework was constructed according to the design knowledge requirements and conceptual product design process.Then,the design knowledge extraction model based on joint learning was proposed for overlapping entity and relation extraction.This model used the ELECTRA algorithm for text semantic coding,and converted triple extraction to mapping two entities based on joint learning.The results of the comparison experiment on real design-related data and the DuIE dataset showed better performance of the proposed design knowledge extraction model.A creative cultural design knowledge graph for conceptual product design was developed to provide designers with relevant,reasonable visual design knowledge and further support conceptual product design.

Key words: conceptual product design, design knowledge graph, entity and relation extraction, joint learning, industrial design

摘要： 为提高产品概念设计过程中设计知识获取效率,解决设计知识碎片化、推理性差、可视化程度不高等问题,提出了基于联合学习的设计知识抽取与图谱构建方法。首先,根据设计知识需求和产品概念设计流程,构建了面向产品概念设计的设计知识图谱框架,然后,分析了设计知识提取过程中实体关系重叠问题,建立了基于联合学习的设计知识抽取模型,该模型通过ELECTRA算法进行语义编码,利用联合学习方法将实体关系抽取转换为各关系上两实体映射。 通过真实设计数据和DuIE公共数据集对比实验表明,该模型知识抽取效果显著优于对比模型。最后,将该方法应用于文化创意产品设计知识图谱开发,为设计师提供关联化、可推理、可视化的设计知识,辅助进行产品概念设计。

关键词: 产品概念设计, 设计知识图谱, 实体关系抽取, 联合学习, 工业设计