Research On Ontology Alignment Based On Word Embedding

The rapid development of computer networks has led to explosive growth of data.Knowledge graphs provide an efficient way for the representation,organization,management,and utilization of heterogeneous,massive,and dynamic data on these networks,and increase the level of network intelligence.Ontology is a conceptual model abstracted from the objective world.It contains concepts and the relationships between concepts.It is an abstraction of knowledge and a formal expression of concepts in a certain field and their relationships.The purpose of constructing ontology is to describe domain knowledge.The parallel ontology development phenomenon in the domain raises the requirement of ontology alignment.Ontology alignment takes the ontology as input and the alignment result as output to determine the correspondence between semantically related entities in different ontologies,which is the key contributor to the interoperability of the semantic web.With the development and maturity of representation learning technology in the field of natural language processing,experts and scholars have begun to try to apply it to the ontology alignment problem.Compared with the original feature extraction method,representation learning technology has certain advantages.The embedding of words pre-trained with a large number of related corpora can represent the inherent semantic information of the words,and the recognition literals represent different words with the same semantic meaning.However,the generation of word embedding depends on the context of the word,which leads to the inability of word embedding to accurately distinguish semantic similarity and description relevance.In order to solve this problem,this paper combines the specific tasks of ontology alignment,uses SCBOW model and Knowledge distillation model to improve word embedding,and uses the improved word embedding to obtain entity embedding,thereby calculating the similarity of two entities and obtaining candidate entity pairs..Considering that using word embedding to complete the ontology alignmenttask only utilizes the semantic information of the entity,and the ontology is a network of entities organized in hierarchical relationships and contains rich structural information.In this paper,we use the MTrans E model to embed two ontology and ontology space mapping to get the structure embedding of the entity,and then use the structure embedding of the ontology to calculate the similarity of the entity pair to obtain the final alignment result.Experiments show that both improved word embedding and structural embedding of added entities have improved the accuracy of similarity calculation of conceptual entities to a certain extent.