Research On Domain Ontology Assisted Construction Based On Knowledge Representation Learning

In order to complete many different types of tasks,researchers have created a knowledge base in this related knowledge area.However,the data in different knowledge bases shows the characteristics of loose structure and multi-source heterogeneity.In order to solve this problem,researchers began to create a method that satisfies the structured description of the concept of the relevant field-"Ontology" for a specific field.The process of creating a domain ontology is extremely complicated.In this process,not only the interchange of various methods but also the participation of human resources is required.To optimize its process,the purpose of this article is to use related automatic learning methods When the domain ontology is created,it performs a certain auxiliary role for a specific link in the process.Its specific tasks are: given the domain knowledge base,in the process of domain ontology creation,using related automation methods to assist the mapping of entities in the ontology construction process to the corresponding concept classes,that is,clustering the entities.For this task,this study processes the data extracted from the Sem Med DB knowledge base,uses a multi-hop strategy to complete the corresponding entity matching,and constructs a knowledge base for the diabetes field through triplet filtering.Based on this knowledge base,experiments were conducted to compare the effectiveness of the models used in this paper.The key to completing the auxiliary construction requirements in this study is how to effectively represent the data in the knowledge base-entities,and the organization of the knowledge base in this study is the knowledge graph,which is expressed as a multi-relationship graph.Therefore,in this study,we use the Trans X translation model,Rotat E and other complex space models in knowledge representation learning to represent the feature of the entity,and propose a direct relationship model to represent the feature of the entity from the relational semantic space in the knowledge base,and then further use The fusion model of direct relationship and knowledge representation learning represents the entities in the diabetes domain knowledge base,and uses it to complete the clustering task of entities in the ontology creation process.In this study,two indicators,standard mutual information and contour coefficient,are used to evaluate the clustering results.At the same time,visualization tools are used to display and analyze the entity feature vectors learned by the model and the clustering effect.The experimental results show that the fusion model using direct relationship and knowledge representation learning can obtain the best results in the research,in which: NMI value reaches 0.503 and Mean S value reaches 0.475.Explain that through the direct relationship and knowledge representation learning fusion model for entity feature learning,the results can help domain experts in the concept of the ontology creation process.