Quaternion-based three-dimensional rotation knowledge graph embedding method

Abstract

The invention discloses a quaternion-based three-dimensional rotation knowledge graph embedding method. The method comprises the following steps of: 1, defining three relation modes and related formalized expressions of a knowledge graph; 2, defining the basis of the quaternion and three-dimensional rotation representation based on quaternion multiplication; and 3, carrying out relation three-dimensional rotation modeling based on quaternion, deducing the modeling capability of the model for three relation modes of the knowledge graph, and carrying out model training to obtain better quaternion vector representation. Quaternions are introduced to carry out modeling representation on entities and relationships in the knowledge graph, the entities are represented as a group of vectors in a three-dimensional space, and the relationships represent three-dimensional rotation transformation between the entities, so that the problem that a current embedded model cannot effectively learn and infer various relationship modes, especially combination relation modes, in the knowledge graph can be solved. Therefore, the fuzzy degree of the entity and the relationship can be effectively reduced, and the knowledge graph link prediction accuracy is improved.

Description

technical field [0001] The invention relates to the field of knowledge graphs, in particular to a quaternion-based three-dimensional rotation knowledge graph embedding method. Background technique [0002] Google proposed the concept of Knowledge Graph in 2012, aiming to use knowledge graph to enhance the function of its search engine. The knowledge graph is a carrier used to store objective factual information, and describes real-world entities, concepts, and events as well as the relationship between them in a graph. Entities are the points of the directed graph, and relationships are the edges of the directed graph. Each piece of knowledge is expressed as a triplet (h, r, t), where h represents the head entity, t represents the tail entity, and r represents the head and tail Relationships between entities. With its powerful semantic computing capabilities and open interconnection capabilities, the knowledge graph plays an important role in the fields of recommendation s...

AI Technical Summary

Problems solved by technology

[0007] In order to overcome the above-mentioned defects in the prior art, the present invention provides a quaternion-based three-dimensional rotation knowledge graph embedding method, which introduces quaternion to model and represent the entities and relationships in the knowledge graph, and the entity Represented as a set of vectors in three-dimensional space, the relationship represents the three-dimensional rotation transformation between entities, which can solve the problem that the current embedded model cannot effectively learn and reason about various relational patterns in the knowledge graph, especially the combination of relational patterns, so that it can effectively Reduce the ambiguity of entities and relationships, and improve the accuracy of knowledge graph link prediction

Images