Research On Ontology Learning Method In Patent Field

Information resources has become one of the most strategic resources in modernsociety with the rapid development of information technology. It has been one of themost concerned hot issues in modern businesses, governments and the masses about howto retrieve the required contents from the massive information resources and acquireinformation from these contents. As a knowledge organization way, ontology has manyrelative technologies to be researched and has important significance and value ontheoretical research and practical application, such as extracting hyponymy relationshipsbetween ontology concepts, inducing semantic hierarchy of ontology concepts, theconstruction approach and practical usage of the constructed ontology in real systems.Ontology concept is one of the basic resources in an ontology and the hyponymyrelationship between concepts is one kind of the basic semantic relationships. The qualityof the hyponymy relationships has significant influence to the constructed ontology. Inpatent documents,“Technical Theme” is a summary of the invention and its usage fields.Thus, constructing a “Technical Theme” cored ontology in patent domain is the basis toachieve patent analysis work such as hotspot theme monitoring. For those purposesmentioned above, this paper aims its work at those “Technical Theme” concepts obtainedfrom patent documents in New Energy Vehicle domain. In this paper, we implementedthe automatic extraction of hyponymy relationships between these concepts, andimproved the quality of the hyponymy relationship set by constructing a SemanticHierarchical Graph of Concepts, and eliminate those redundancy relationships andconfliction relationships. Furthermore, we constructed a patent ontology in New EnergyVehicle domain by the basic of the constructed Semantic Hierarchical Graph of Concepts,and then we integrate the ontology into the patent knowledge platform as a knowledgerepository to provide patent knowledge services. The major works in this paper are listedas follows:(1) This thesis proposes a concept semantic hierarchy induction approach. First, weanalyze those word-formation disciplines during the formalization process of technologytheme concepts in patent domain. Based on this, a relative decoration based approach andan association rule based approach are proposed to hierarchical relation extraction afterabstract modeling these concept terms. Second, we improved the current association rulebased approach by analyzing the statistical discipline between hyponymy relationships. The experimental result shows that the approach of hyponymy relation extraction canachieve high accuracy and recall rate, the approach we proposed can achieve satisfiedelimination result. The result of the experiment proves the validation of the approachproposed in this paper.(2) This thesis proposes the Semantic Hierarchical Graph of Concepts and a relationelimination approach according to the redundancies and conflictions in the graph. Byanalyzing the multi-path phenomenon in the graph, this approach eliminates redundantrelations by traversing the redundancies. Further, by comparing the confidence ofrelations and the proposed relation effect approach between relations in the detectedcycles in the graph, conflicting relations are eliminated. Experiment in the patents obtainsF-values of0.965on redundancy elimination and0.893on confliction elimination. Theexperiment result proves the efficiency and active significance of this approach.(3) This thesis proposes a patent ontology construction approach in New EnergyVehicle domain. Based on the constructed Semantic Hierarchical Graph of Concepts, weuse the existing ontology editing tool to construct the frame ontology by obtaining theconcepts set, semantic relationship set and the axiom set. Then we fill the knowledge intothe frame by the ontology developing tool to construct the patent ontology. After that, weevaluate and tested the logical reasoning capability of the ontology by the ontologyevaluation tool and the ontology reasoning plugins. The result proves that the constructedontology can achieve its usability and application value.(4) This thesis designs and implements the patent ontology based Patent KnowledgePlatform. The major purpose of construction a patent ontology is to apply it to theplatform and provide semantic services to the patent information retrieval systems, andthen make the platform more intelligent and able to provide various services to usersprecisely. By abstracting and modeling the functions in the platform, we design thepatent ontology based platform by the basic of the constructed abstract model. Finally,we develop the platform with the software developing technology and deployed it to thereal devices. By presenting the running results and introducing the opened APIs and theirusages in the platform, we explained the features in the platform in more detail.