融合概念和属性信息的领域知识图谱补全方法

Domain knowledge graph completion method incorporating concept and attribute information

针对领域知识图谱具有严格的模式层和丰富的属性信息的特点,提出一种融合概念和属性信息的领域知识图谱补全方法.首先对领域知识图谱模式层中的概念使用可建模语义分层结构的HAKE模型进行嵌入表示,建立基于概念的实例向量表示;其次对数据层的实例三元组和属性三元组进行区分,通过注意力机制对实例的属性和概念进行融合,建立基于属性的实例向量表示;最后对基于概念和基于属性的实例向量表示进行联合训练以实现对实例三元组的评分.使用基于DWY100K数据集构建的知识图谱、MED-BBK-9K医疗知识图谱和根据某钢铁企业设备故障诊断数据构建的知识图谱进行实验,结果表明所提出方法在领域知识图谱补全中的性能优于现有知识图谱补全方法.

Aiming at the characteristics of domain knowledge graphs with strict schema layers and rich attribute information,a method of domain knowledge graph completion incorporating concept and attribute information is proposed.Firstly,the concepts in the schema layer of the domain knowledge graph are represented by embedding using the HAKE model which can model semantic hierarchical structures to build a concept-based instance vector representation.Then,a distinction is made between instance triples and attribute triples for the data layer,and an attribute-based instance vector representation is obtained by incorporating the attributes and concepts of the instance through the attention mechanism.Finally,the concept-based and attribute-based instance vector representations are jointly trained to achieve scoring of the instance triples.Experiments are conducted using the knowledge graph constructed based on the DWY100K dataset,the medical knowledge graph MED-BBK-9K and the knowledge graph constructed based on equipment fault diagnosis data of a steel enterprise,and the experimental results show that the performance of the proposed method in domain knowledge graph completion is better than the existing knowledge graph completion methods.