Entity alignment method and device based on edge type attention mechanism

Abstract

The invention discloses an entity alignment method based on an edge type attention mechanism. The entity alignment method comprises the following steps: constructing a type dual graph of an electric power knowledge graph; training node characteristics of the type dual graph by adopting an attribute attention mechanism and obtaining embedding of each node in the type dual graph; converting nodes inthe type dual graph into attention coefficients of relationships in the knowledge graph, gathering neighbor entities and relationship information by the nodes in the knowledge graph according to theattention coefficients of the relationships, and obtaining a final structured embedding result; calculating the value of the loss function and aligning entities in different knowledge graphs to obtainentity pairs aligned by entities in different knowledge graphs. The invention also discloses a method for realizing the entity alignment method based on the edge type attention mechanism. According to the method, the thought of relationship information mining is broadened, the structural characteristics of entities are mined, and the method is high in accuracy, good in practicability, simple andreliable.

Description

technical field [0001] The invention belongs to the field of power system big data processing, and in particular relates to an entity alignment method and device based on an edge type attention mechanism. Background technique [0002] In the task of building a large-scale knowledge base, it is necessary to process a large amount of entity data from multi-source knowledge bases, which are usually stored in the knowledge base in the form of knowledge graphs, such as figure 1 shown. Therefore, at the beginning of building a knowledge base, it is first necessary to establish a knowledge fusion method to fuse multi-source knowledge graphs. Due to the different sources of information in different knowledge bases, as well as the differences in manual definition and proofreading, semantically identical entities will have different representations in different knowledge bases. Entities with the same name may represent semantically the same thing, or two things. Therefore, before k...

AI Technical Summary

Problems solved by technology

However, previous methods treat neighbor nodes equally, so they cannot perceive the weight of the relationship, nor can they guarantee the consistency of the same type of relationship in the training process.

The second is that the relationship attributes have not been fully utilized, and the aligned nodes are also aligned with the surrounding relationships. Effective use of relationship attribute information can obtain better entity embedding, such as in figure 2 Among them, the relationship type, unit, and material connected to the Chinese entity resistor, and the relationship type (type), unit (unit), and material (material) connected to the English entity resistance are aligned

[0005] In summary, current KG entity alignment methods ignore the role of relation type information in entity alignment

Images