Knowledge graph representation learning method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of machine learning, and discloses a knowledge graph representation learning method and device, electronic equipment and a storage medium, which can improve the accuracy of vector representation in a knowledge graph. The method specifically comprises: clustering entities in the knowledge graph to obtain at least two entity clusters; determining the distance between any two entities in the knowledge graph based on the vector representation of the entities; obtaining a score function of triads based on triads describing relationships between entities, each triad including a vector representation of a head entity, a vector representation of a tail entity, and a vector representation of a relationship; on the basis of the distance between the entities, the at least two entity clusters and the scoring function, constructing an objective function, wherein the objective function is in positive correlation with the distance between the two entities belonging to the same entity cluster, and the objective function is in negative correlation with the distance between the two entities belonging to different entity clusters; and updating vector representations of entities and relationships in the knowledge graph by minimizing the objective function.

Description

technical field [0001] The present invention relates to the field of machine learning, in particular to a knowledge map representation learning method, device, electronic equipment and storage medium. Background technique [0002] The vectorized representation of entities and relationships in knowledge graphs is a basic and important technology in knowledge graph technology, and it is also the basis for knowledge graphs to perform subsequent tasks including knowledge reasoning, question answering, and recommendation. The key problem of representation learning is to learn low-dimensional distributed embedding vectors of entities and relations, which are generally targeted by scoring functions (distance-based and similarity-based). [0003] In the prior art, CTransR and TransF are representative for solving one-to-many and many-to-many problems. Among them, CTransR refers to Cluster-basedTransR. For each specific relationship r, CTransR first performs AP clustering (h, t) acc...

AI Technical Summary

Problems solved by technology

[0005] However, the vector representation of the knowledge graph obtained by the existing technology cannot distinguish two different tail entities obtained from the same head entity and the same relationship, resulting in no

Images