《计算机应用》唯一官方网站 ›› 2024, Vol. 44 ›› Issue (11): 3371-3378.DOI: 10.11772/j.issn.1001-9081.2023111677

• 人工智能 • 上一篇    下一篇

融合实体语义及结构信息的知识图谱推理

王利琴1,2,3, 张特1, 许智宏1,2,3, 董永峰1,2,3(), 杨国伟4   

  1. 1.河北工业大学 人工智能与数据科学学院,天津 300401
    2.河北省大数据计算重点实验室 (河北工业大学),天津 300401
    3.河北省数据驱动工业智能工程研究中心 (河北工业大学),天津 300401
    4.天津市公安局北辰分局,天津 300401
  • 收稿日期:2023-12-05 修回日期:2024-04-26 接受日期:2024-05-11 发布日期:2024-05-30 出版日期:2024-11-10
  • 通讯作者: 董永峰
  • 作者简介:王利琴(1980—),女,河北张家口人,高级实验师,博士,CCF会员,主要研究方向:智能信息处理、知识图谱
    张特(1997—),女,河北石家庄人,硕士,主要研究方向:智能信息处理、知识图谱
    许智宏(1970—),女,河北张家口人,教授,博士,CCF会员,主要研究方向:智能搜索、知识图谱
    杨国伟(1976—),男,天津人,主要研究方向:大数据、区块链。
  • 基金资助:
    河北省高等学校科学技术研究项目(ZD2022082);河北省高等教育教学改革研究与实践项目(2022GJJG049);中国高等教育学会2022年度高等教育科学研究规划课题(22XX0401);河北省研究生教育教学改革研究项目(YJG2023023)

Fusing entity semantic and structural information for knowledge graph reasoning

Linqin WANG1,2,3, Te ZHANG1, Zhihong XU1,2,3, Yongfeng DONG1,2,3(), Guowei YANG4   

  1. 1.School of Artificial Intelligence,Hebei University of Technology,Tianjin 300401,China
    2.Hebei Data Driven Industrial Intelligent Engineering Research Center (Hebei University of Technology),Tianjin 300401,China
    3.Hebei Engineering Research Center of Data?Driven Industrial (Intelligence Hebei University of Technology),Tianjin 300401,China
    4.Beichen Branch of Tianjin Municipal Public Security Bureau,Tianjin 300401,China
  • Received:2023-12-05 Revised:2024-04-26 Accepted:2024-05-11 Online:2024-05-30 Published:2024-11-10
  • Contact: Yongfeng DONG
  • About author:ZHANG Te, born in 1997, M. S. Her research interests include intelligent information processing, knowledge graph.
    XU Zhihong, born in 1970, Ph. D., professor. Her research interests include intelligent search, knowledge graph.
    YANG Guowei, born in 1976. His research interests include big data, blockchain.
    First author contact:WANG Liqin, born in 1980, Ph. D., senior experimentalist. Her research interests include intelligent information processing, knowledge graph.
  • Supported by:
    Science and Technology Research Project of Higher Education Institutions in Hebei Province(ZD2022082);Research and Practice Project of Teaching Reform of Higher Education in Hebei Province(2022GJJG049);Higher Education Scientific Research Planning Project of China Association of Higher Education for the Year of 2022(22XX0401);Teaching Reform Research Project of Postgraduate Education in Hebei Province(YJG2023023)

摘要:

目前,图注意力网络(GAT)通过引入注意力机制对目标实体的邻域实体赋予不同权重并进行信息聚合,使得它更关注实体的局部邻域,忽略了图结构中实体和关系之间的拓扑结构;而且在多头注意力后将输出嵌入向量简单拼接或平均,导致注意力头之间相互独立,未能捕捉不同注意力头的重要语义信息。针对GAT应用于知识图谱(KG)推理任务时未充分挖掘实体结构信息和语义信息的问题,提出融合实体语义及结构信息的知识图谱推理(FESSI)模型。首先,使用TransE将实体和关系表示为同一空间的嵌入向量。其次,提出交互注意力机制,将GAT中多头注意力重新融合成多个混合注意力,增强注意力头之间的交互性,以提取目标实体更丰富的语义信息;同时,利用关系图卷积网络(R-GCN)提取实体的结构信息,并通过权重矩阵学习GAT和R-GCN的输出特征向量。最后,使用ConvKB作为解码器进行评分。在知识图谱数据集Kinship、NELL-995和FB15K-237上的实验结果表明,FESSI模型的效果优于多数对比模型,在3个数据集的平均倒数排名(MRR)指标上的结果分别为0.964、0.565和0.562。

关键词: 知识图谱, 知识图谱推理, 关系图卷积网络, 图注意力网络, 交互注意力机制

Abstract:

Currently, Graph ATtention network (GAT) assigns different weights to entities in the neighbourhood of the target entity and performs information aggregation by introducing an attention mechanism, which makes it pay more attention to the local neighbourhood of the entity and ignore the topology between entities and relations in the graph structure. Moreover, the output embedding vectors are simply spliced or averaged after the multi-head attention, resulting in the independence of attention heads, and fails to capture important semantic information of different attention heads. Aiming at the problems that GAT does not fully mine entity structural information and semantic information when it is applied to knowledge graph reasoning task, a Fusing Entity Semantic and Structural Information for knowledge graph reasoning (FESSI) model was proposed. Firstly, TransE was used to represent entities and relationships as embedding vectors in the same space. Secondly, an interactive attention mechanism was proposed to reintegrate the multi-head attention in GAT into multiple hybrid attentions, which enhanced the interaction between the attention heads to extract richer semantic information of the target entity. At the same time, the structural information of the entity was extracted by utilizing the Relational Graph Convolutional Network (R-GCN), and the output feature vectors of GAT and R-GCN were learned through weight matrices. Finally, ConvKB was used as a decoder for scoring. Experimental results on the knowledge graph datasets Kinship, NELL-995 and FB15K-237 show that the FESSI model outperforms most comparison models, with the Mean Reciprocal Rank (MRR) index on the three datasets of 0.964, 0.565 and 0.562, respectively.

Key words: knowledge graph, knowledge graph reasoning, Relational Graph Convolutional Network (R-GCN), Graph ATtention network (GAT), interactive attention mechanism

中图分类号: