| With the rapid development of the Internet, the informationization degree becomesmore and more high in the world. But now the World Wide Web service cannot makepeople feel very satisfied, the problem mainly reflects that degree of intelligence is nothigh enough in the search engine. An emerging way for searching data solutions-semanticweb, as one of the solutions, well embodies the advantages of the intelligent. In order tobuild the semantic web, it’s necessary to build the resource description framework. For theChinese, its job is to find the SVO ingredients of a sentence, and the triple willautomatically build the semantic web.Dependency parsing is a syntactic analysis technique, which consists of the languagemodel, Eisner algorithm and the maximum spanning tree algorithm. Through to finddepends on the predicate of several chunks, further to find the subject and the object of thesentence. By this way, simple SVO elements of sentences can be obtained within a shortertime. On this basis, semantic role labeling is a manifestation of shallow semantic analysis.Through combining each of the chunks in the sentence to analyze that which role they playin the sentence. Using this method to analyze the sentence elements has a higher accuracy.Feature selection is an important part of the two methods, it has a great influence on thefinal results. This paper focuses on the features of the selection, choose the appropriatefeature to make the result better. The system using the trained language models for featureselection and matching makes the two methods collocation to complete extraction of SVOtriples. The system combines two methods and improves the simple semantic role methodsbrought about a matter of time.The method is tested on the public unlabeled corpus, and results compare with thatobtains separate semantic role labeling and dependency grammar analysis methods. Thecorrect rate spends higher than that of dependency analysis method and the method lesstime than semantic role labeling. |
PDF Full Text Request