Research Of Ontology Population Based On Linked Open Data

In recent years,with ontology technology developing rapidly,many relatively perfect domain ontologies have appeared in many fields and have achieved good application.However,there are ontologies in most areas,which are relatively lack of knowledge of the instance.For the users of the metallic materials ontology,what they want is not only that the rich and perfect data in schema is contained in the ontology,but also that there is a large number of instance knowledge in this domain.There are a large number of linked open data(LOD),such as DBpedia and Wikipedia,which have many data that can be used to fill the ontology,and these LOD contain various domains.However,there are a lot of inconveniences in using these LOD directly.Because of the above problems,this paper proposes to populate the existing domain ontology with LOD,which can not only classify the data in LOD according to the field,but also enrich the knowledge of existing domain ontology.Meanwhile,taking the field of metallic materials as an example,the data of metallic materials in LOD is extracted and added to the existing domain of metallic materials.The following research work has been done in the paper:(1)Extracting specific domain data from LOD(such as DBpedia)based on domain ontology.Domain ontology,Wikipedia and topic model algorithm are used to extract the entrance of extracting specific domain data from DBpedia.The direct link subgraph semantic distance algorithm is improved and a similarity strategy is designed to obtain domain specific knowledge in the DBpedia.(2)Based on machine learning algorithms,the domain specific data in LOD is added to existing domain ontology.Firstly,the domain specific population data in LOD is extracted by the filling requirements of domain ontology,and the chain triple is generated according to the population data.Then,according to the used machine learning algorithm,a conversion strategy about chain triple feature is designed.Finally,population positions in domain ontology are got by using the machine learning algorithm for each chain triple.In addition,in this paper,CRF algorithm in probability graph model algorithm and classification algorithms(such as the logistic regression algorithm,random forest algorithm and SVM algorithm)are used to obtain the chain triples' population positions in the domain ontology.(3)Taking the metallic material field as an example,the rationality and validity of the method are analyzed and evaluated by conducting experiments.In addition,for domain knowledge extraction algorithm and ontology population strategy,the time performance and F1-measure are used to evaluate the performance of these algorithms.The experimental results show that the designed algorithms are feasible,and the time performance is acceptable.At the same time,a prototype system is designed to visually display the extracted metallic materials knowledge.And a ontology population system is designed to show the whole ontology population process in detail.