Knowledge graph embedding method based on attribute aggregation and storage medium thereof

Abstract

The invention discloses a knowledge graph embedding method based on attribute aggregation and a storage medium thereof. The method comprises the following steps: aggregating attributes in a knowledgegraph and converting the aggregated attributes into entities; all positive examples are formed by utilizing triple groups existing in a knowledge graph, a positive example training data set O+ is formed, a value domain class inverse example and a relation class inverse example are constructed, and an inverse example training data set O- is formed; and establishing a knowledge graph embedded objective function, and solving the objective function by using the training data formed in the previous step. According to the method, attributes are converted into entities through attribute aggregation,and then the knowledge graph is embedded, so that loss of attribute information in a knowledge graph embedding result is prevented. According to the method, a value domain type inverse case construction method based on a relation value domain and a relation type inverse case construction method based on a type relation domain are adopted, so that the probability of occurrence of false inverse cases is reduced, the quality of a training sample is improved, and the consistency of a knowledge graph embedding result and a knowledge graph real structure is indirectly improved.

Description

technical field [0001] The present invention relates to the field of knowledge graphs, in particular to a knowledge graph embedding method for attribute aggregation in knowledge graphs, converting attributes in knowledge graphs into entities, and increasing the amount of information that can be contained in entity embedding and relationship embedding in knowledge graphs and its storage media. Background technique [0002] The knowledge graph has the ability to describe complex relationships in the real world. Since its concept was proposed in 2012, it has received extensive attention from academia and various application fields. Today, there are already a large number of knowledge graph systems, which have played an important role in the fields of information extraction, expert systems, knowledge question answering, and social network analysis. [0003] A knowledge graph is a graph structure that describes entities and the relationships between entities. A knowledge graph ...

AI Technical Summary

Problems solved by technology

Network information has the characteristics of complex sources and poor credibility

Using Web corpus as the basis will increase the workload of knowledge map embedding, and also introduce uncertainty into the process of knowledge map embedding, making it difficult to guarantee the credibility of knowledge map embedding results

[0008] (2) When embedding the knowledge graph, the attributes of the entity are not considered, and a large amount of information is lost. Therefore, the embedding results formed are difficult to apply in the field of entity attribute discovery, which has limitations.

[0011] (1) A completely random replacement method is used in the generation of negative examples in the training samples, and the result may produce false negative examples (for example, "Yao Ming's nationality is China" is a randomly generated false negative example), thus affecting the final embedding result

[0012] (2) When embedding knowledge graphs, this method also does not consider the attributes of entities and loses a lot of information. Therefore, the embedding results formed are difficult to apply in the field of entity attribute discovery, which has limitations.

[0013] Therefore, it can be seen that most of the existing research and inventions on knowledge graph embedding do not consider the attribute information of entities, making it difficult to apply the embedding results of knowledge graphs in the field of entity attribute discovery.

[0014] In addition, in the existing research and invention of knowledge map embedding, the randomization method used in the generation of counterexamples in training samples may produce false counterexamples and affect the credibility of knowledge map embedding results

Images