Frame Semantic Based Answer Sentences Extraction For Chinese Reading Comprehension In College Entrance Examination

At present,most of the knowledge on the Internet is presented in the form of text.And the knowledge representation and in-depth understanding of Internet shows the level of intelligent information processing to some extent.QA system can verify the ability of internet to understand the text to some extent.Comprehensive Questions is an important branch of QA system,which can be used as an effective means to study and evaluate questions.What’s more,it can also be used to extract answers.Based on the National863 project,aiming at the comprehensive questions of Chinese Scientific Articles,this thesis puts forward to extract answers by aid of frame semantic collocation,relationship and view.During this process,the author will lead into manifold ranking model to spread scores by analyzing the semantic relevance between sentences.The purpose of it is to select the higher score of Top-4 as the sentences.The main research contents and results are as follows:Firstly,Chinese frame semantic knowledge is used in the extraction of answers.Based on the cognitive theory,frame network formally describes the relationship between syntax and semantics,which uses real corpus as the research background.It is different from grammatical relation and discourse relation.There are two merits of application of frame semantics.On the one hand,frame network has much more information than general dictionaries.On the other hand,it reveals the relationship between semantic fields in reading materials rather than that between adjacent sentences.Secondly,manifold ranking model is adopted when ranking the answers.This model is a semi supervised global ranking algorithm,which makes the most of the semantic relevance of the frame between the candidate sentences.It is to make nodes have reasonable ranking scores through iteration propagation until it reaches steady state.Experiment shows that the localization of questions not only extends or completes the semantic field but also greatly improves the correctness when extracting candidate answers.This thesis contains three parts.Firstly,it is an overview of comprehensive questions in the college entrance examination.Then,it introduces Chinese frame semantics in detail.And on this basis,it proposes the computational method about sentence semantic relativity and constructive method about text semantic view.Lastly,it applies these methods into authentic Chinese reading comprehensive questions and evaluates it at the same time.