Method Of National College Entrance Examnation History Problems Answer Generation Based On Retrival

With the development and popularization of the Internet,a huge amount of information has been generated on the Internet.Human beings need to obtain information in a simpler and faster way.Search engines have gradually failed to meet human's requirements for information acquisition.The automatic question answering system is a more advanced form of information service than the search engines.The system returns to the user no longer a list of documents sorted according to relevance,but a more accurate natural language answer.The rapid development of artificial intelligence technology has greatly promoted the automatic question answering system,and also promoted the promotion of the automatic question answering system in more subdivided fields.For example,in recent years,foreign research institutions have begun to study automatic question-and-answer technology in the field of examination,and hope to develop humanlike intelligent robots with certain logical thinking and reasoning ability,such as the basic education examination robot project in the United States.In our country,IFLYTEK has taken the lead in organizing the college entrance examination answering robot project.This paper hopes to construct an automatic answering system for the college entrance examination history short answer questions,and focus on the application of the answer generation method based on the search in the answering system.The main research contents of this paper include:Test analysis and knowledge base analysis.Through the statistical analysis of the college entrance examination history short answer questions,the different classifications of the questions and the characteristics of each category are summarized.A certain amount of historical corpus is obtained through the network and other channels.The knowledge coverage of the knowledge points in the knowledge base is analyzed in the college entrance examination history simulation test.Problem and material keyword extraction.In order to extract keywords more accurately,this paper combines a variety of methods,using supervised learning algorithms to extract keywords in the problem;for the material in the historical short answer questions,this article combines the characteristics of historical questions,using the method of part-of-speech annotation to extract the key word.Candidate answer extraction and sorting.In this paper,based on the keyword information of the problem and the material,the relevant documents are retrieved from the knowledge base,and the document is cut by the sliding window method to finally generate the candidate answer.In order to complete the sorting of the candidate answers,this paper preliminarily trained the sentence encoding model based on recurrent neural network on the large-scale corpus,and then embedded the problem,material and candidate answers into the semantic space by using the pre-trained sentence encoding model,and in the semantic space,uses questions and materials to sort the candidate answers.System performance evaluation.In this paper,each module is evaluated separately,and then the overall performance of the system is manually evaluated.Through the experiment of 95 sets of college entrance examination history short-answer questions for nine sets of questions,the problem-solving method of this paper is 5% higher than the benchmark method.