Method Of History Short-Answer Problems Based On Background Material

With the continuous development of the Internet technology,the way of access to information is undergoing profound changes,transforming from the search engine to the currently Q&A system.The Q&A system makes the information more straightforward,and it does not return the relevant page collection just like the search engine,but returns the answer to the user's question directly.Artificial intelligence technologies are rapidly evolving and different types of question-answering systems have been spawned according to the needs of different industries.In recent years,foreign research institutions begun to apply natural language processing technology to the field of examination,hoping to develop a logical thinking and reasoning ability of humanoid intelligent robot,such as the Todai project that Japan launched and hoped to pass the university entrance examination.In China,IFLYTEK also organized the college entrance examination answer robot project.Based on the work of predecessors,this paper hopes to build an automatic problem solving system for senior history short-answer questions,and focuses on the application of background materials in the problem solving system.In this paper,based on the characteristics of the senior history short-answer problems based on background material,a fine-grained automatic problem solving solution is proposed.By classifying the types of the senior history short-answer problems and summarizing the characteristics of each type,this paper concludes the difficulties of each type of problems.Combining the types of questions in the problems and the interdependent syntactic structures of different questions,this paper decomposes the compound questions into simple questions using natural language processing technology,and then uses the unsupervised method to rank the importance of the words in the question and then extract the keywords of the question.By discussing the relationship between question and background material,this paper deeply analyzes the key information types in the background material,and proposes a two-stage strategy-the generation of candidate words and the classification of candidate words for keyword extraction of the material.Based on intention keywords of the question and background materials,this paper searches related knowledge documents in history knowledge base and generates candidate answer sets from these knowledge documents.The paper also tried unsupervised sorting method,supervised learning method and convolution neural network matching model for the ranking of candidate answers,and get the answer by deducting and combining the multiple candidate answers with high relevance.This paper builds a large-scale knowledge base containing senior history textbooks,Baidu encyclopedia on historical domain and Chinese Wikipedia,and collects a large number of history past and mock problems of college entrance examination as experimental data from the Internet.Through the evaluation experiment of 350 senior history short-answer problems,this paper can get an average ROUGE-1 recall rate of 0.374,which is 0.07 higher than the best benchmark method.