Training method, alignment method, device and equipment of knowledge graph alignment model

Abstract

The application discloses a training method, alignment method, device and equipment for a knowledge map alignment model, which belong to the field of artificial intelligence and cloud technology. The method includes: obtaining a first entity pair set, the first entity pair set includes a plurality of first entity pairs that are not labeled with alignment results; based on the predicted alignment probabilities of each first entity pair, selecting a plurality of first entity pairs from each first entity pair The first candidate entity pair; calculate the difficult alignment degree of each first candidate entity pair; based on the difficult alignment degree of each first candidate entity pair, select a plurality of first target entity pairs from each first candidate entity pair; obtain each The label alignment result of the first target entity pair is obtained according to the predicted alignment probability and label alignment result of each first target entity pair to obtain a target knowledge map alignment model. This application passes two screenings, which greatly reduces the number of entity pairs that need to be labeled, saves time and cost, improves the training speed of the model, and improves the alignment efficiency.

Description

Technical field [0001] Embodiments of the present application relate to artificial intelligence and cloud technology, and in particular, to a training method, alignment method, apparatus, and equipment of knowledge map alignment model. Background technique [0002] The knowledge map is composed of an interconnected entity and the relationship between the entities, and the knowledge map alignment technology is of great significance to construct a large-scale high quality knowledge map. Typically, the entities in the two knowledge maps are aligned based on the training well-trained knowledge map, in order to implement the contents of other knowledge maps in one knowledge map, thereby integrating knowledge maps of different vertical domains of different particles. [0003] In the relevant technique, two knowledge maps are obtained in advance, and any of the entities in one of a knowledge map and any of the other knowledge maps, manually conducting the two entities, according to this...

AI Technical Summary

Problems solved by technology

[0004] In the above method, the number of training data sets is very large, and manual labeling of each entity pair in the training data set takes a lot of time, which affects the training speed of the model, thereby affecting the alignment efficiency

Images