摘要
知识图谱嵌入将实体和关系表示投影到低维向量空间,从而简化计算、保留图谱结构信息,以便进行知识图谱补全,知识图谱问答等下游任务。而大多数融合附加信息的知识图谱嵌入模型只关注实体描述、文本描述等外部信息,没有考虑到知识图谱中关系对头实体、尾实体的不同类型约束。因此,提出一种基于双重关系类型约束的知识图谱嵌入模型,针对头尾实体的领域范围不同,设计了双重关系类型约束,考虑关系对同一实体处于不同位置的不同影响。首先,从三元组结构中提取双重关系类型约束特征;然后,压缩双重关系类型约束特征维度;最后,联合训练双重关系类型约束特征和三元组结构表示,以提高实体和关系表示的准确性。链接预测实验结果表明,在FB15K-237数据集上,相较于最优的基准模型,所提模型的Hit@10和Hit@1提高了1.5%,倒数平均排名提高了1.4%。
Knowledge graph embeddings project entities and relational representations into low-dimensional vector space,thereby simplifying computation and preserving graph structure information for downstream tasks such as knowledge graph completion and knowledge graph question answering.However,most knowledge graph embeddings models that fuse additional information only focus on external information such as entity description and text description,and do not take into account the different types of constraints on head and tail bodies in knowledge graphs.Therefore,a knowledge graph embeddings model based on dual relationship type constraints is proposed,according to the different domain scope of the head and tail entities,a dual relationship type constraint is designed to consider the different effects of the relationship on the same entity in different positions.First,the double relation type constraint feature is extracted from the triple structure,then the double relation type constraint feature dimension is compressed,and finally the double relation type constraint feature and the triple structure representation are jointly trained to improve the accuracy of the entity and relation representation.The link prediction experimental results show that on the FB15K-237 data set,compared with the optimal baseline model,the Hit@10 and Hit@1 of the proposed model are improved by 1.5%,and the MRR(Mean Reciprocal Rank)is improved by 1.4%.
作者
季文龙
余敦辉
张万山
JI Wenlong;YU Dunhui;ZHANG Wanshan(School of Computer Science and Information Engineering,Hubei University,Wuhan 430062,China;Hubei Key Laboratory of Big Data Intelligent Analysis and Application(Hubei University),Wuhan 430062,China;Engineering and Technical Research Center of Hubei Province in Software Engineering,Wuhan 430062,China)
出处
《微电子学与计算机》
2025年第8期29-37,共9页
Microelectronics & Computer
基金
国家自然科学基金(61977021)
国家科技创新2030-重大项目(2020AAA0107700)。
关键词
知识图谱
知识图谱嵌入
联合训练
双重关系类型约束
特征降维
knowledge graph
knowledge graph embedding
joint training
dual relationship type constraint
feature dimensionality reduction
