Research On Approaches Of Mining Axioms For RDF Data

With the rapid growth of Semantic Web in several decades,there is an exponential growth in the scale of linked data on the web.Ontology is now palying an important role in Semantic Web.In order to describe some domain knowledge precisely,we usually need a set of terminologies with specific meanings.Such meanings and the relationships of these terminologies can be marked by a set of axioms.However,a lot of large-scale RDF datasets have generous facts but no corresponding axioms to describe how the classes and properties are defined、described and associated with each other.For instance,approximately 16%properties in DBpedia dataset with version of 2015 lacks the declaration of a rdfs:domain value and other characteristics about properties and classes such as disjointness of classes and properties are not stated at all.Hence,how to mine axioms from instance level to enrich ontology knowledge base has been one of the most important issuses for the current Semsntic Web research field.In order to solve the problem mentioned above,we propose two approaches to mine axioms based both on OWL 2 and Patterns.The primary study in this paper are described as following parts:1)Do a research on mining axioms under the restrictions of the three profiles of OWL 2.We first use SPARQL query language to get instances information from RDF datasets after which we can construct transaction tables.Then,based on these transaction tables,we apply the association rule mining algorithm to mine axioms that satisfies the restrictions of OWL 2.2)Do a research on mining axioms based on patterns.We extract and normalize the axiom patterns from several well-known ontology repositories which are frequently used for empirical experimentation.Then,we construct the axioms dependency tree and use the dependency of the nodes to get candidate axioms of each pattern.Finally,we compute t.he score of each candidate axiom and compare the score with a pre-set threshold to judge if the candidate axiom has the potential to be an axiom.3)The proposed approaches are implemented and tested on large-scale datasets.We use manual evaluations、threshold analysis to show the results of the evaluations.We also explain some of the evaluation results and present specific examples.The result of the experiments shows that our approach can make an effective axiom mining on large-scale datasets and play a positive role for the enrichment of ontology knowledge base.