Research On Geography Question Analysis In Gaokao

In current society,with the rapid development of technology,researches in the field of artificial intelligence is becoming more and more in-depth.These studies make computer play a good performance in the various intellectual activities that people par-ticipation in.Therefore,people are not satisfied with the current status,constantly trying to challenge the new field of artificial intelligence.At present,a very popular artificial intelligence challenge is the automatic question answering system,in partic-ular,let the computer through a variety of different levels of entrance examinations.Question answering system has broad prospects in the field of artificial intelligence and natural language processing.Although the question and answer system has undergone rapid development,but because of the particularity of the examination question text,the current mainstream automatic question answering system is generally solving the factoid problems,and college entrance examination questions are usually in the form of multiple choice ques-tions or short answer questions.The question analysis is the basic part of the question answering system,only by correct and comprehensive analysis the question,extract-ing the key information of the question,the computer is able to answer the question correctly.At present,general question analysis and semantic parsing models and algo-rithms can not be applied to such data,so it is necessary to further improve the semantic parsing model for this kind of data.In this paper,we improve the current semantic parsing model for the question text.Firstly,we built a corpus of geography college entrance examination questions,and through summarizing the characteristics of test questions by artificial observation,we present a new model of problem-based type-argument meta-template understanding model for Beijing geography college entrance examination questions.The model clas-sifies the form of the problem and determines its key components for each individual type,makes up of the template form,which is convenient for the downstream modules to solve the problem.Second,we model automatic template generation and imple-ment test questions to template conversion method based on sequence labeling model.Finally,in order to improve the correct rate of conversion,we achieve a rule-based au-tomatic test questions to template conversion based on Chinese sentence pattern theory.The experiment result shows that the proposed type-argument meta-template model is better than the traditional semantic parsing model,which is not too trivial and can keep the key information completely to improve the efficiency of the question solving.The automatic template generation based on the statistics method can further improve the efficiency of template conversion.Compare to the statistics method,the tule-based template generation method can further collect the sentence and syntactic structure in-formation of the test questions,thus improving the correct rate of template conversion.