基于多特征提取和对比学习的知识图谱链接预测  

Knowledge graph link prediction based on multi-feature extraction and contrastive learning

作　　者：李华昱[1] 李海洋 王翠翠[1] 满笑军 Li Huayu;Li Haiyang;Wang Cuicui;Man Xiaojun(College of Computer Science&Technology,China University of Petroleum(East China),Qingdao Shandong 266580,China)

机构地区：[1]中国石油大学(华东)计算机科学与技术学院,山东青岛266580

出　　处：《计算机应用研究》2025年第2期530-538,共9页Application Research of Computers

基　　金：山东省自然科学基金资助项目(ZR2020MF140);中国石油大学(华东)研究生创新基金资助项目(22CX04035A)。

摘　　要：针对传统知识图谱链接预测方法提取图谱节点特征角度单一,且在训练过程中较少考虑节点间复杂的交互作用,构建的负例三元组质量较低等问题,提出了一种链接预测方法,旨在充分利用知识图谱节点间的相互作用和图结构蕴含的交互信息,考虑从多特征角度识别出三元组中的缺失事实。首先,通过不同的节点特征提取方式从不同角度获得节点的嵌入表示,并聚合邻居节点特征以增强其实体语义信息;其次,用多个卷积操作提取实体和关系之间的全局关系和过渡特征,通过深度特征提取的方式处理实体和关系的信息交互;最后,通过引入对比学习,干预负例三元组的构建,同时增强负例三元组的特征,提高所构建三元组的质量,最终通过计算余弦相似度筛选出预测实体。实验结果表明,提出的方法在知识图谱链接预测任务中的多个评价指标相比对比模型均有提高,同时验证了所提方法在处理多关系的复杂知识图谱时的有效性。Aiming to address issues such as the single perspective of traditional knowledge graph link prediction methods,limi-ted consideration of complex interactions between nodes during training,and the low quality of constructed negative triplets,this paper proposed a novel link prediction approach.This method aimed to fully utilize the interactions between nodes in the knowledge graph and the interactive information implied by the graph structure,considering the identification of missing facts in triplets from multiple feature perspectives.Firstly,it obtained embedded representations of nodes from different perspectives through various node feature extraction methods and aggregated neighboring node features to enhance their entity semantic information.Secondly,multiple convolution operations were employed to extract global relationships and transitional features between entities and relations,handling the interaction of entity and relation information through deep feature extraction.Lastly,by introducing contrastive learning,it intervened in the construction of negative triplets,simultaneously enhancing the features of negative triplets to improve the quality of constructed triplets.Finally,it filtered out predictive entities by calculating cosine similarity.The experimental results show that the proposed method improves several evaluation metrics in the knowledge graph link prediction task compared to the comparison model,and also verifies the effectiveness of the proposed method in dealing with complex knowledge graphs with multiple relationships.

关 键 词：知识图谱 链接预测 图结构 对比学习 负采样 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]