Transform-based relation mode adaptive contrast learning knowledge graph embedding method

The invention discloses a relationship mode adaptive contrast learning knowledge graph embedding method based on Transform, and belongs to the technical field of knowledge graphs. According to the knowledge graph triple data, the knowledge graph is converted into the relation path, the direct relati...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: WANG JUN, ZHANG GUIXU, XIAO MENGJIN, LIU SANNYUYA, WANG SHIJIN, GAN JIANHOU, WANG MINGJIE, CHEN KEN
Format: Patent
Sprache:chi ; eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The invention discloses a relationship mode adaptive contrast learning knowledge graph embedding method based on Transform, and belongs to the technical field of knowledge graphs. According to the knowledge graph triple data, the knowledge graph is converted into the relation path, the direct relation between entities is captured, potential semantic association and indirect interaction between the entities are revealed, and the problem that a Transform model cannot directly process knowledge graph data is solved; according to the method, anchor point samples are generated according to different relation modes in a relation path, negative samples are constructed by using a negative sampling method of entity or relation replacement, and contrast loss is constructed to capture richer relation information; according to the method, the relation path is masked, so that the capturing capability of the Transform model on the implicit relation in the knowledge graph is enhanced, and the trained Transform model can eff