Time sequence knowledge graph completion method based on space-time architecture

Abstract

The invention discloses a time sequence knowledge graph completion method based on a space-time architecture, and the method is characterized in that the method comprises the following steps: dividinga to-be-completed time sequence knowledge graph into a plurality of static knowledge sets according to the time marks of knowledge of the to-be-completed time sequence knowledge graph, respectively constructing a plurality of knowledge networks through the knowledge in each set, and obtaining a plurality of snapshots; constructing a polyhedral graph attention network, inputting the snapshots intothe polyhedral graph attention network, and obtaining static embedded representation of the entity under each snapshot; constructing a self-adaptive time sequence attention mechanism, and obtaining afinal embedded representation of the entity according to the static embedded representation of the entity and by using the self-adaptive time sequence attention mechanism; and calculating confidenceof knowledge in the to-be-completed time sequence knowledge graph through final embedding representation of the entity, and predicting missing content in the to-be-completed time sequence knowledge graph through the confidence. The invention has higher expansibility and flexibility, and can be combined with any static completion method to complete the timing sequence knowledge graph.

Description

technical field [0001] The invention belongs to the technical field of knowledge graphs, and in particular relates to a time-series knowledge graph complement method based on a spatio-temporal framework. Background technique [0002] As a dynamic knowledge base system, time-series knowledge graph has attracted much attention in recent years because of its good application value and considerable application prospects in many practical fields. However, due to the limitations of extraction methods and the complexity of data sources, the existing large-scale time-series knowledge graphs still face serious incompleteness problems, and a large amount of knowledge is missing. Therefore, the temporal knowledge graph completion task aimed at predicting missing knowledge in temporal knowledge graphs has become an important research topic in this field in recent years. [0003] The static knowledge graph embedding representation technology aims to map the elements in the knowledge gra...

AI Technical Summary

Problems solved by technology

[0004] Although the existing work has achieved good results in the task of time-series knowledge graph completion, there are two obvious deficiencies in the existing work: (1) First, the existing work regards the time-series knowledge graph as a set of independent knowledge , and then process each knowledge independently and learn the corresponding embedding representation for the elements in each knowledge separately

(2) Second, most of the existing work deals with knowledge at different moments independently, which leads to the inability of existing work to capture the correlation between knowledge at different times

In fact, there is a close causal relationship between knowledge at different times, ignoring the temporal correlation between knowledge makes these models unable to effectively utilize past relevant knowledge and make accurate predictions for current missing knowledge

Therefore, the performance of existing representation learning-based temporal knowledge graph completion methods is far from satisfactory.

Images