Knowledge graph pre-training method based on structured context information

Abstract

The invention discloses a knowledge graph pre-training method based on structured context information, and the method comprises the steps: constructing an instance composed of context triples for a target triple, employing a triple integration module to code each context triple of the instance, and obtaining an integration vector; forming a context vector sequence by the integrated vectors of allthe context triples of the instance, and encoding the context vector sequence by adopting a structured information module to obtain structural representation vectors of the triples; and calculating the structure representation vector of the triple by adopting a universal task module to obtain a label prediction value of the triple, and updating the structure representation vector of the triple based on the cross entropy loss of the label prediction value of the triple and the label true value until the training is finished to obtain an optimized structure representation vector of the target triple. The structural representation vector of the triple obtained by the method combines contextual information.

Description

technical field [0001] The invention belongs to the technical field of data storage and processing, and in particular relates to a knowledge graph pre-training method based on structured context information. Background technique [0002] The Knowledge Graph can be regarded as a directed labeled graph, and the facts in the graph are represented as triples in the form of (head entity, relationship, tail entity), which are abbreviated as (h, r, t ). In recent years, knowledge graphs have developed rapidly in terms of construction and application, and have broad application prospects in artificial intelligence fields such as semantic search, information extraction, and question answering. [0003] Since the graph structure in knowledge graph contains a large amount of valuable information, it is crucial to extract deep structural information for various knowledge graph tasks, such as entity typing, link prediction, entity alignment, etc. The representation learning method embe...

AI Technical Summary

Problems solved by technology

Therefore, the main challenges of adapting pre-training KG representation learning models for different tasks are: (1) Regardless of the specific KG downstream tasks, the pre-trained model should be able to automatically capture the deep structural contextual information of a given triple; (2) The representation of entities and relationships needs to be trained in different ways according to different downstream tasks and different structural features of the input data of downstream tasks to improve its robustness

Images