Knowledge graph completion method based on meta-relation learning

• Main Authors: ZHOU YINGHAI, TIAN ZHIHONG, SUN YANBIN, XU TIANFU, QIU RIXUAN, CUI YU, REN YITONG, LI MOHAN, ZHENG, ZHIBIN, JIANG YUNXIN, LIU YUAN, QIU JING, LU HUI, WANG RUI, HE QUN
• Format: Patent
• Language: Chinese; English
• Published: 07.05.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention provides a mapping knowledge domain completion method based on meta-relation learning, which comprises the following steps of: performing embedded representation learning based on mapping knowledge domain data to obtain static embedded representations of entities and relations in a mapping knowledge domain; performing meta-task training according to the static embedding representation to obtain learned parameters, and performing meta-task testing on the learned parameters to select meta-relation learning parameters; and fusing the static embedded representation and the meta-relation learning parameters to obtain a comprehensive entity representation, calculating a relation path by using the comprehensive entity representation, updating the relation path by using a graph neural network to obtain an updated representation of nodes on the relation path, and generating a final path representation, and performing inter-entity relationship prediction by using the path representation to obtain a predic