Knowledge graph representation learning training-oriented local training method

Abstract

The invention discloses a knowledge graph representation learning training-oriented local training method. The method comprises the following steps of S1, obtaining knowledge graph training data; S2,calculating a vector space dimension which should be used for training; S3, judging whether a new model can be obtained by adjusting the original model or not; S4, carrying out model training; and S5,model and data storage. According to the method, the vector space dimension required for training can be calculated according to the data scale of the knowledge graph, so problems of poor model performance caused by too small dimension and waste of calculation resources and time required for training caused by too large dimension are avoided. For the changed knowledge graph, training adjustment can be performed on the basis of the original representation learning model, so the representation learning model of the changed knowledge graph is quickly obtained, a large amount of time required bytraining is saved, and an upper-layer application can perceive the change of the knowledge graph more timely.

Description

technical field [0001] The invention belongs to the technical field of computer knowledge graphs, and relates to a local training method for representation learning and training of knowledge graphs, in particular to a dynamic representation learning training method considering the continuous change of knowledge graphs. Background technique [0002] A knowledge graph is a graph composed of entities as points and relationships between entities as edges. Since Google launched its first version of knowledge graph in 2012, knowledge graph has developed rapidly. Knowledge graphs can provide query capabilities and reasoning capabilities, and have been applied in many fields such as social networks, finance, medical care, anti-fraud, and corporate relationship divisions. However, due to the high complexity of the graph search algorithm, when the amount of data in the graph continues to increase to a certain extent, the computational efficiency in the graph will decrease, and the pr...

AI Technical Summary

Problems solved by technology

For the original knowledge, it means that the learning model has been learned in the training of the map data before the change, while the traditional method cannot use the learned knowledge in the training process of the map after the change, but starts from scratch. study again

This makes the system need to learn the same knowledge repeatedly every time, which causes a waste of computing resources and an unacceptable time cost

Images