一种面向中医医案知识图谱的链路预测模型

A Link Prediction Model for Knowledge Graph of TCM Medical Records

知识图谱有助于实现智能化、个性化的中医药知识服务,其中链路预测可以解决知识图谱中缺失信息的发现与还原,也是目前知识图谱应用领域中的研究热点之一。但目前补全的医学知识图谱很少覆盖类型和层级结构,而且链路预测未考虑到关系三元组。本文创新性地提出一种新的归纳推理模型HSTP(HierarchicalStructureType),基于类型和层级结构获取信息,利用知识图谱中实体之间的语义关联,将类型、节点与关系进行融合,然后提出一个关系相关网络来学习不同模式对归纳链路预测的重要性,最终得到真实和完整的中医医案知识图谱。结果表明,该模型能够有效表达实体之间的语义关联,在链路预测任务的基准数据集上提高了3.9%左右,可以为解决知识图谱中缺失信息的发现与还原提供研究基础。

Knowledge graph(KG)helps to realize intelligent and personalized knowledge service of TCM.Link prediction can solve the discovery and restoration of missing information in KG,and it is also one of the research hotspots in the field of KG application.However,the current complementary medical KG rarely covers types and hierarchical structures,and link prediction does not take relational triples into account.This study proposed a new inductive reasoning model HSTP(Hierarchical Structure Type).The information is obtained based on the type and hierarchical structure,and the semantic association between entities in the KG is used to integrate the type,node and relationship.Then,a relational network is proposed to learn the importance of different models for inductive link prediction,and finally a real and complete KG of TCM medical records is obtained.The results showed that the model could effectively express the semantic association between entities,and improve about 3.9%on the benchmark dataset of link prediction task,which provided a research basis for solving the discovery and restoration of missing information in KG.