Knowledge graph fusion method based on entity sequence coding

Abstract

The invention discloses a knowledge graph fusion method based on entity sequence coding. The method comprises the steps of 1, knowledge graph entity representation learning; 2, selecting a path code and an alignment model; 3, performing cross-language entity alignment on the model, wherein in the source language knowledge graph space, a two-hop sequence corresponding to other seed entities is constructed for one entity, a sequence possibly corresponding to the entity is constructed in the target language knowledge graph space, an alignment sequence with the highest probability is found out, and then a node at the same position is found out from the alignment sequence to serve as an alignment node of the node; 4, adding a new candidate seed node. Aiming at the problem of insufficient training corpus of a deep learning model in the prior art, the method based on entity path representation learning is put forward.

Description

technical field [0001] The present invention relates to the technical field of knowledge map application, in particular to a knowledge map fusion method based on entity sequence coding. Background technique [0002] At present, well-known Internet companies at home and abroad, such as Google, Baidu, Tencent, and Microsoft, have built their own knowledge bases. These knowledge bases provide a large amount of knowledge service information. For example, Google's knowledge base system Knowledge Vault has stored 1.6 billion items information, and more information is currently being collected. The application service of Baidu's knowledge graph has increased by 160 times in 5 years. These companies use more semantic information that knowledge graphs can provide to provide more intelligent search services and provide Internet users with portable services. [0003] These knowledge graphs contain a large amount of common sense information, and the integration of these cross-language...

AI Technical Summary

Problems solved by technology

Using deep learning to learn node vectors, label known alignment entity pairs, and train alignment models for alignment are current research hotspots. This method performs alignment operations based on the semantic information of entities and the relationship semantics between entities, improving the The accuracy of the model, but there are problems with this type of model, and it takes a lot of manpower to label the training corpus

[0004] Moreover, due to the good performance of the deep learning model in the entity alignment task, but the lack of training corpus, if the training corpus is insufficient, it is impossible to learn an accurate and efficient alignment model

Images