融合实体描述与路径信息的知识图谱表示学习模型  

Knowledge graph representation learning model combining entity description and path information

作　　者：李军怀[1,2] 武允文 王怀军[1,2] 李志超 徐江[3] LI Junhuai;WU Yunwen;WANG Huaijun;LI Zhichao;XU Jiang(Collaborative Innovation Center of Modern Equipment Green Manufacturing in Shaanxi Province,Xi’an University of Technology,Xi’an 710048,China;School of Computer Science and Engineering,Xi’an University of Technology,Xi’an 710048,China;China National Heavy Machinery Research Institute Co.,Ltd.,Xi’an 710032,China)

机构地区：[1]西安理工大学陕西省现代装备绿色制造协同创新中心,陕西西安710048 [2]西安理工大学计算机科学与工程学院,陕西西安710048 [3]中国重型机械研究院股份公司,陕西西安710032

出　　处：《智能系统学报》2023年第1期153-161,共9页CAAI Transactions on Intelligent Systems

基　　金：国家重点研发计划项目(2018YFB1703000);陕西省现代装备绿色制造协同中心自主研发或开放基金项目(112-256092104)。

摘　　要：知识图谱表示学习方法是将知识图谱中的实体和关系通过特定规则表示成一个多维向量的过程。现有表示学习方法多用于解决单跳知识图谱问答任务,其多跳推理能力无法满足实际需求,为提升多跳推理能力,提出一种融合实体描述与路径信息的知识图谱表示学习模型。首先通过预训练语言模型RoBERTa得到融合实体描述的实体、关系表示学习向量;其次利用OPTransE将知识图谱转化成融入有序关系路径信息的向量。最后构建总能量函数,将针对实体描述和路径信息的向量进行融合。通过实验分析与对比该模型在链路预测任务上与主流知识图谱表示学习模型的性能,验证了该模型的可行性与有效性。Knowledge graph representation learning is a process of representing knowledge graph entities and relations in a multidimensional vector through specific rules. Existing representation learning methods are mostly used to solve the single-hop knowledge graph question-and-answer task, but their multi-hop reasoning ability cannot meet the actual demand. To improve the multi-hop reasoning ability, a knowledge graph representation learning model combining entity description and path information is proposed. First, the learning vector of entity and relation representation is obtained using the pre-training language model RoBERTa. Second, OPTransE is used to transform the knowledge graph into a vector integrating the path information of an ordered relation. Finally, the total energy function is constructed to fuse the vectors of entity description and path information. The feasibility and validity of the model are verified by comparing its performance in a link prediction task with that of the mainstream knowledge graph representation learning model.

关 键 词：知识图谱 表示学习 多维向量 多跳推理能力 实体描述 路径信息 能量函数 向量融合 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]