融合实体描述与关系路径信息的知识图谱嵌入模型  

Knowledge Graph Embedding Model Fusing Entity Description and Relational Path Information

作　　者：翟社平 李方怡[1] 亢鑫年 杨锐 ZHAI Sheping;LI Fangyi;KANG Xinnian;YANG Rui(School of Computer Science and Technology,Xi′an University of Posts and Telecommunications,Xi′an 710121,China;Shaanxi Key Laboratory of Network Data Analysis and Intelligent Processing,Xi′an University of Posts and Telecommunications,Xi′an 710121,China)

机构地区：[1]西安邮电大学计算机学院,西安710121 [2]西安邮电大学陕西省网络数据分析与智能处理重点实验室,西安710121

出　　处：《小型微型计算机系统》2024年第12期2867-2874,共8页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61373116)资助;工业和信息化部通信软科学项目(2018-R-26)资助;陕西省教育厅科学研究计划项目(18JK0697)资助;陕西省社会科学基金项目(2016N008)资助;陕西省重点研发计划项目(2022GY-038)资助;西安市社会科学规划基金项目(17X63)资助;西安邮电大学研究生创新基金项目(CXJJYL2021044)资助;陕西省大学生创新创业训练计划项目省级项目(202211664053)资助;陕西省大学生创新创业训练计划项目省级项目(202211664086)资助。

摘　　要：目前已有的知识图谱嵌入方法主要关注三元组结构信息,未充分利用外部文本补充知识图谱的语义信息,同时由于现有模型仅使用单跳输出结果进行知识图谱嵌入,这导致在嵌入过程中丢失了大量的路径信息.针对以上问题,本文提出了一种融合实体描述与关系路径信息的语义嵌入模型TPKGE,该模型首先利用软注意力机制对三元组和实体描述文本进行过滤,提取与特定关系相关的有效信息,其次以分层的方式学习多跳关系路径特征嵌入,最终将两部分结果合并到同一实体和关系嵌入表示中.在4个公开的数据集上进行实验,结果表明TPKGE与其他基线模型相比更具竞争力,验证了同时考虑文本描述以及多跳关系路径信息的有效性.The existing knowledge graph embedding methods mainly focus on triplet structure information,and do not make full use of external text to supplement the semantic information of knowledge graph.At the same time,due to existing models only use single-hop output results for knowledge graph embedding,a lot of path information is lost in the embedding process.To solve the above problems,this paper proposes a semantic embedding model TPKGE that integrates entity description and relational path information.The model first utilizes a soft attention mechanism to filter triplets and entity description texts,extracting effective information related to specific relationships,and then learns multi-hop relational path feature embedding in a hierarchical way.Finally,the two parts of the result are merged into the same entity and relational embedding representation.Experiments on four publicly available datasets show that TPKGE is more competitive than other baseline models,verifying the effectiveness of considering both text description and multi-hop relationship path information.

关 键 词：知识图谱嵌入 文本描述 关系路径 注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]