Materialization Based On Multi-step Ontology Reasoning

In recent years,big data and artificial intelligence have developed rapidly.When faced with massive data and highly expressive ontology,the efficiency of reasoning is greatly challenged.Converting reason problems into query problems through materialization is one of the mainstream methods of current query answering,but materialization has limitations on the reliability and finiteness of results.Ontology-mediated querying(OMQ)is a core reasoning task in this paper.Compared with the cases of the plain databases,where query answering is based on the records explicitly declared in tables,OMQ is defined using the certain answer semantics over possibly infinitely many models of possibly infinite sizes.Therefore,OMQ is in general much more challenging.In this paper,we adopt a pure materialization-based approach and solve the problem of the infinite growth of the materialized database by the depth of the query.This method can guarantee the soundness and completeness of the results under DL-LiteNhorn.In other cases,the approximation can be as close as possible to the completeness while ensuring the soundness of the result.This paper develops techniques of generating and managing the model of materialization and implements them in a prototype gOWL.From a system engineering perspective,the materialization allows us to design a modular architecture to integrate off-the-shelf efficient SPARQL query engines.Within gOWL,we build two OMQ systems gOWL-3X and gOWL-AD by employing RDF-3X and TriAD(a distributed engine),respectively.The preliminary encouraging experiments show that gOWL outperforms PAGOdA,Ontop,and Pellet(with speedup up to three orders of magnitude)both in realistic datasets(DBpedia)and synthetic datasets(LUBM and UOBM)In summary,this paper proposes a multi-step materialized method and framework for query answering to deal with weak reasoning ability and poor efficiency,which effectively improves the efficiency and the completeness of the results.It will provide new ideas for applying ontology reasoning to big data.