A Hybrid Method For Hypernym Relation Extraction And Ontology Learning

With the rapid development of the Internet,more and more applications generate a large amount of user data and texts.In these data and texts,we can often obtain very valuable information or laws,which can further create value for society and enterprises.Through Knowledge Graph,we can further organize and integrate the huge amount of Internet data,and an excellent Knowledge Graph is very important for knowledge recommendation,query understanding and personalized search.The core task of building a Knowledge Graph is Knowledge Extraction.Hypernymy Relation,a fundamental component of a classification system,is used to describe the "is-a" relationship between two concepts,which plays a key role in many Natural Language Processing tasks.These relations can serve as a basis for building more complex structures or as effective background knowledge for many word comprehension applications,therefore,an effective approach to obtain hypernymy relations is very important.Many scholars and experts have achieved outstanding results in this field,among which the better-known relation extraction methods are pattern based and distributed approach,however,pattern-based approach has higher accuracy but lower coverage,on the other hand,distributed approach can have higher coverage but requires manual data generation,which is often impractical in the case of large data volume.To solve the above problems and to automate the process of extracting hypernym relations,in this paper,we propose a hybrid method based on pattern-based approach and distributed approach SHP-ML,which effectively overcomes the shortcomings of the above two single methods and can perform automatic hypernym relations extraction given "seed" concepts.This approach can generate a taxonomy tree of the extracted concepts.In order to be user-friendly,this paper designs and implements a visualized automatic hypernym relation extraction system based on the proposed hybrid method.