Method And System For Solving Geographical Multiple-choice Questions In High Schools

With the rapid development of Question Answering(QA)technology,it is be-coming a hot research topic that developing machines to pass human examinations.This thesis focuses on solving geographical multiple-choice questions in high schools.Although some QA approaches have been proposed for various application scenarios,these approaches are not suitable for solving geographical multiple-choice questions in high schools because of the difficult of question understanding,construction of domain-specific knowledge base and knowledge inference.The main work and contributions of this thesis are as follows:1.This thesis proposes a framework of solving geographical multiple-choice ques-tions in high schools,including semantic understanding of multiple-choice ques-tion,judgement of options and integration of answers.A question understanding model based on two-level semantic templates is adapted to convert multiple-choice questions,including the stem and option of questions,into 11 templates by trigger words and typical sentence patterns.In order to judge the option of questions,3 kinds of methods are proposed,including a method based on textual semantic sim-ilarity,a method based on retrieving domain-specific knowledge base and methods based on specific question types with two sub methods:a method based on numer-ical calculation and a method based on matching.A method based on the priority of judgement methods and a method based on logistic regression are proposed to integrate answers given by different methods for different options.2.This thesis proposes a approach based on the structure of the table and typical sen-tence patterns to build the geographical relation template knowledge base.3.Based on these methods,a complete system of solving multiple-choice questions in geographical Gaokao,XiaoLi,was designed and implemented.The experiment shows that the method based on textual semantic similarity had a high coverage and an ordinary performance,while the rest of the judgement methods achieved extremely high accuracy but poor coverage.However,these methods showed good complementarity in the experiment.The former integration reached 39.09%score and the latter one reached 42.73%in the experiment,which outperformed any single method.In a formal examination environment,XiaoLi tested 5 unopened exam papers and achieved 43.63%accuracy in average,which indicates that XiaoLi has a preliminary ability of solving geographical multiple-choice questions in high schools.