Research On Named Entity Recognition And Relation Extraction Facing To Domain-oriented Knowledge Base Construction

Any information processing system is inseparable from the support of data andknowledge base, especially natural language processing system. As humans, weneed to accumulate a large number of the necessary knowledge to understand thenatural language, so does a computer system. So a high quality knowledge base is anecessity for a natural language processing system. Although there are severaldifferent choices for the task of knowledge extraction considering of thetechnical detail and reliability, it generally includes two common and closely relatedsubtasks: named entity recognition and relation extraction. Our study mainlyinvolves the following aspects:(1) A semi-supervised named entity recognition method based on multi-patternfusion is proposed and the corresponding experiments and results analysis aresupplied.(2) Study the method of building knowledge base based on Chineseencyclopedia. From the aspect of building domain oriented knowledge base, we firstextract the corpus of related domain from Wikipedia and then establish a domainoriented knowledge base by using the Infobox of the Wikipedia.(3) Analyze entity property to extract relevant features by using the corpus weconstructed. For the entity property relation extraction task: Firstly, we conductfeature extraction; Secondly, training two different classifiers for entity propertyrelation extraction respectively based on maximum entropy and SVM models;Thirdly, the classifiers we trained are tested on artificial annotated corpus and theexperimental result are given.