Character relationship mining model training method and character relationship mining method and device

Abstract

The invention provides a character relationship mining model training method, a character relationship mining method and a character relationship mining device. The training method comprises the following steps: acquiring a space-time knowledge graph; carrying out random sampling according to a positive sample of the space-time knowledge graph, generating a negative sample, and determining head entity initial embedding, relation initial embedding, tail entity initial embedding and time embedding of the positive sample and the negative sample; performing vector rotation on the initial embedding of the head entity and the initial embedding of the tail entity to obtain quaternion embedding of the head entity and quaternion embedding of the tail entity; respectively replacing the initial embedding of the head entity and the initial embedding of the tail entity with corresponding quaternion embedding of the head entity and quaternion embedding of the tail entity to obtain a processed positive sample and a processed negative sample; and iteratively training the character relationship mining model to convergence by adopting the processed positive sample and the processed negative sample. According to the technical scheme, the relationship between entities evolved along with time change can be mined, and the integrity of the knowledge graph is improved.

Description

technical field [0001] The present invention relates to the technical field of knowledge graphs, in particular to a training method for a character relationship mining model, a character relationship mining method and a device. Background technique [0002] Knowledge graph is a series of different graphs showing knowledge development process and structural relationship, using visualization technology to describe knowledge resources and their carriers, mining, analyzing, constructing, drawing and displaying knowledge and their interconnections. Among them, the spatio-temporal knowledge map is to add spatio-temporal attribute information to entities, which can realize the management and query of spatio-temporal data in specific target fields. The spatio-temporal knowledge graph includes a large number of facts containing spatio-temporal information, and each fact constitutes a quadruple (s, r, o, t), which includes the head entity s, relation r, tail entity o and time t. The ...

AI Technical Summary

Problems solved by technology

The space-time knowledge graph usually has the problem of missing links, which limits the application of the spatio-temporal knowledge graph. more complete to reveal interrelationships between entities

[0003] At present, the knowledge map embedding method is often used to complete the knowledge map, and predict the unknown entity and the relationship between the entities based on the known entity and the relationship between the entities. These methods can pay attention to the static relationship between entities, but many entities. The relationship between entities is not static. For example, many relationships between characters will evolve over time, and the existing knowledge map completion methods cannot predict the evolution of relationships between entities over time.

Images