Research On Semantic Information Extraction Based On Predicate-Argument Structure

With the development of the Internet, the amount of information on the Internetis also rapid growth in index form, such a flood of Internet data has extremelyimportant value. However, the Internet data specific to its own heterogeneity,randomness and non-semantic features, and it is very difficultto find the datawhichyou need, so it is very urgent to find a way which computer can quickly andaccurately understand the natural language.The most fundamental way to understand natural language is semanticinformation extraction which can transform a flat text into semantic structure. Thetraditional way of semantic information extraction is firstly transforming into textparsing grammatical structure, grammatical structure and then establish mappingrelationships directly with the semantic structure. This approach has the drawbacks,transformingsyntax into semanticsdirectly span relatively large, so it is difficult toachieve.In this paper it made a research on shallow semantic analysis of the text, and thenproposed a way of semantic information extraction which base onpredicate-argument structure (PA structure). This method adds PA structure betweengrammatical structure and semantic structure. PA structure is a cross between thegrammatical structure and semantic structure; it has both grammatical features andshallow semantic features.So with the participation of the PA structure, semanticinformation extraction will greatly ease the span traditional extraction methodproblems.After the introduction of the PA structure, semantic information extraction is divided into two steps: using the syntactic structure to get the PA structure andmapped the PA structure to the semantic structure. As semantic role labelingandsemantic role inductionare relatively mature, so getting the PA structure is relativelysimple, the article does not dwell on it. In the article we use ontology as a guide tomap PA structure and semantic structure.By selecting the appropriate form of thebackground ontology, then eleven established the PA structure with backgroundknowledge ontology mapping relationship and thereby achieve the semanticinformation extraction.In Chapter4, it details the core algorithm of the semantic information extractionbased on PA structure. And in Chapter5, the article did experiments on the algorithmby usingPropBank, FrameNet and SemLink as experimental data. At the same timethe article make deep analysis on the results.