Dependency Syntax Analysis Based On Question Answering System

With the rapid development of Internet and technological advances, today the intelligent search has gradually become the mainstream. Most Internet users search answers by way of natural language on the web and brief and accurate answers are provided to the users. The requirement for higher ability to understand natural language(such as Chinese) indicates that the computer must understand not only machine codes, but also the user’s language. This has lead to many high-tech products, such as chat robot. This feature determines the answering system can better meet the requirements of the user’s query, it can provide users with a more accurate answer. Making computer has the equivalent ability of understanding with a person is a leap of science and technology, which has driven the rapid development to artificial intelligence in the field of semantic understanding, such as the online intelligent systems of Zhihu and Baidu, intelligent psychological consulting system and intelligent chat robot, etc..Question answering(QA) system usually contains three major modules. These are analysis module, information retrieval module and answer extraction module. This thesis focuses on the analysis modules, which is the main module in a QA system analysis. Syntactic analysis can discover problems in the analysis, which has huge advantages. And it can be more profound and more methodical in the analysis and the problem of induction.In this thesis, the maximum spanning tree discriminative dependency analysis method is based on dependency analysis method and deterministic dependency parsing method(also known as the analysis method of parsers) complementary relationship, which puts forward the maximum spanning tree dependency analysis method, and determine the type dependency analysis method to combine the dependence analysis method. Niver’s model modified maximum generation dependency analysis is used in the results of model and tree to the weights of the edges, and then search the maximum spanning tree as a dependency tree.Experiment results indicate that,the two new dependency algorithms improve the two evaluation standards(dependency accuracy and sentence accuracy).