Trans_isA:一种基于实体属性和语义层次的表示学习方法  

Trans_isA:a representation learning method based on entity attributes and semantic levels

作　　者：李天宇 王艳娜[1] 周子力[1] 赵晓函 LI Tianyu;WANG Yanna;ZHOU Zili;ZHAO Xiaohan(School of Cyber Science and Engineering,Qufu Normal University,273165,Qufu,Shandong,PRC)

机构地区：[1]曲阜师范大学网络空间安全学院,山东省曲阜市273165

出　　处：《曲阜师范大学学报（自然科学版）》2022年第2期71-78,共8页Journal of Qufu Normal University(Natural Science)

基　　金：山东省自然科学基金(ZR2020MF149)。

摘　　要：传统的表示学习方法将三元组按照独立的个体进行表示,使得实体和关系的表示向量缺乏语义层次信息.同一关系(如is_A关系)链接的两个实体,其自身属性以及语义范畴都可能是不相同的,因此在对is_A关系三元组进行表示学习时,要将头实体和尾实体区分开,以不同的方法进行编码.为此,首先将is_A关系下的实体编码为一个球体,球体间相对位置表示为实体间关系;然后在实体属性的基础上,利用二元关系的传递性判定条件对实体间语义层次进行建模,将其命名为Trans_isA模型.在WN18RR数据集上对Trans_isA进行了链接预测和三元组分类实验,验证了模型的有效性.Traditional representation learning methods represent triples as independent individuals, which makes the representation vectors of entities and relationships lack semantic level information. Two entities linked by the same relationship(such as is_A relationship) may have different attributes and semantic categories. Therefore, we propose the Trans_isA model. Specifically, first encode the entity under the is_A relationship as a sphere, and the relative position between the spheres is expressed as an inter-entity relationship;then, on the basis of entity attributes, the semantic hierarchy between entities is constructed using the transitive judgment conditions of the binary relationship. The link prediction and triple classification experiments of Trans_isA are carried out on WN18 RR datased to verify the validity of the model.

关 键 词：知识图谱 表示学习 实体属性 语义层次 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]