Entity alignment method in four-insurance-one-fund domain knowledge graphs

Abstract

The invention belongs to the technical field of knowledge maps, and particularly relates to an entity alignment method in four-insurance-one-fund domain knowledge graphs. A strategy of dynamically adjusting an embedding loss function according to a relationship type in a TransE model is provided, the TransE loss function under different attributes and relationship conditions is optimized by addingan embedding layer, embedding results of an LSTM network and a Bert to attribute values are fused, semantic information of the attribute values is considered in embedding of the attribute values. Structure embedding and attribute embedding joint training are carried out to finally obtain joint embedding representation of the knowledge graphs. Semantic information and character information of entity embedding are combined, the entity embedding distance and the LCS similarity are combined, and candidate entity pairs are selected under the condition of considering the two similarities. The problem of difficult alignment caused by insufficient link number in the Chinese knowledge graph and inconsistent names of entities in different knowledge graphs can be effectively solved.

Description

technical field [0001] The invention belongs to the technical field of knowledge graphs, and in particular relates to a method for aligning entities in knowledge graphs in the field of four insurances and one fund. Background technique [0002] Since Google proposed the concept of knowledge graphs in 2012, knowledge graphs have developed rapidly, and a number of large-scale knowledge graphs represented by DBpedia, Freebase, Wikidata, YAGO, etc. have emerged; however, the data sources of knowledge graphs are extensive and the data quality is uneven. , leading to the diversity and heterogeneity of different knowledge graphs, there are multiple different entities pointing to the same real-world object in different knowledge graphs, so knowledge fusion has become an important link in the research of knowledge graphs. As a key technology in knowledge fusion, entity alignment, also known as entity matching, is a technique to infer whether different entities from different knowledg...

AI Technical Summary

Problems solved by technology

Matthias[18] and others proposed to align the entities of the knowledge graph with the described text. First, both the text and the knowledge graph are mapped to the vector space through representation learning, and the vector representations of the text and entities are learned, and regularization is added in the learning process. Constraints make the entity vector and description text representing the same entity as close as possible in the vector space, and the representation learning models of knowledge graph and text are trained alternately, so as to achieve alignment without affecting the respective representation learning effects of text and knowledge graph , this method relies on the text information describing the entity, and has poor versatility

Fan et al. [19] proposed the DKRL model modeling triplet in 2017. This model not only considers the energy function based on the relationship representation, but also considers the energy function based on the description information, and proposes two methods for encoding the description information of entities. Method: The continuous bag of words model accumulates the first k keywords of the description information, but this method ignores the word order information, so a method of encoding through a convolutional neural network is proposed

[0013] In the task of large-scale entity matching, due to the imbalance of knowledge graph data, the entity alignment quality of the automatic entity alignment method is limited. Therefore, Zhuang et al. [20] proposed to use the crowdsourcing platform to improve the alignment effect in 2017. First, Partition the knowledge map and perform rough entity alignment, then put aligned entity pairs and unaligned entity pairs into the crowdsourcing platform, and finally establish a partial order relationship between entities to eliminate potential errors and optimize crowdsourcing alignment. Entity effect, but the method still requires a lot of manual operation

Images