摘要
知识图谱是事实三元组的集合,其表示形式为(头实体,关系,尾实体)。为了补全知识图谱中缺失的实体和关系,提出一种基于卷积神经网络的知识图谱补全方法。使用传统嵌入模型训练三元组,得到实体向量和关系向量;将三元组表示成3列矩阵,作为卷积神经网络的输入,卷积后得到三元组的特征表示图;连接所有特征图和权重向量进行点乘得到每个三元组的得分,得分越低证明三元组越正确。实验采用数据集WN18RR、FB15K-237、FB15K分别进行链接预测和三元组分类实验。实验结果表明,与其他方法相比,该方法在Mean Rank和Hit@10指标上都取得了更好的实验结果,证明其可以有效提高三元组预测精度。
Knowledge graph is a collection of factual triples,the representation is(head,relation,tail).In order to complete the missing entities and relations in the knowledge graph,an improved convolutional neural network knowledge graph completion method is proposed.The traditional embedding model was used to train the triples to obtain the entity vector and the relation vector;the triple was represented as a 3-column matrix as the input to the convolutional neural network,and the matrix was convolved with convolution kernel to obtain feature maps;all feature maps were connected,and the weight vector was multiplied to obtain the score of each triple to determine the correctness of a triple.In the experiment,the data sets WN18RR,FB15K-237,FB15K were used to link prediction and triple classification experiments.The experimental results show that compared with other methods,the proposed method achieves better experimental results on the Mean Rank and Hit@10 indicators,which proves that the method can effectively improve the prediction accuracy of the triples.
作者
王维美
陈恒
史一民
李冠宇
Wang Weimei;Chen Heng;Shi Yimin;Li Guanyu(Faculty of Information Science&Technology,Dalian Maritime University,Dalian 116026,Liaoning,China;School of Software,Dalian University of Foreign Languages,Dalian 116044,Liaoning,China)
出处
《计算机应用与软件》
北大核心
2021年第4期250-255,共6页
Computer Applications and Software
基金
国家自然科学基金项目(61371090,61602076,61702072,61976032)
国家社会科学基金项目(15BYY028)
辽宁省自然科学基金项目(20170540232,20170540144,20180540003)
大连外国语大学科研创新团队项目(2016CXTD06)。
关键词
知识图谱
知识图谱补全
卷积神经网络
链接预测
三元组分类
Knowledge graph
Knowledge graph completion
Convolutional neural network
Link prediction
Triple classification