Research On The Solution Method Of Short Answer Questions For Prose Reading Comprehension

Machine reading comprehension has always been a very important research direction in the field of natural language processing.It requires the machine to have the ability of text understanding and reasoning,that is,to answer related questions in combination with given reading materials.Most of the existing research on reading comprehension is carried out on English data.Although many data sets of Chinese reading comprehension tasks have been published,the content and questions of the corpus are more objective and relatively simple.Prose reading comprehension materials are implicit and beautiful,euphemistic and profound,with a wide range of texts and diversified questions.Therefore,the task of prose reading comprehension is very challenging.This paper studies the problem-solving methods of short-answer questions in prose reading comprehension.The main research work is as follows:(1)Prose data set constructionProse has the characteristics of beautiful language,profound artistic conception and long writing.It is also an indispensable part of the Chinese subject of college entrance examination.Through the analysis of the answers to the short answer questions of prose reading comprehension in the college entrance examination,it is found that the key information of the prose reference answers is highly implicit and general.The answer points are generally not less than three points,and sometimes need to be understood in multiple contexts in the text.From the perspective of the answer to the short answer of prose reading comprehension,this paper collects and collates the test questions,simulation questions and real questions of prose reading comprehension in the college entrance examination.The key sentences of the answer are initially marked by the automatic labeling system,and then the results of the initial labeling are manually corrected.The key sentence data set of the answer to the short answer of prose reading comprehension is established.(2)Prose multi-span answer key sentence extraction methodIn the short answer questions of prose reading comprehension,the prose text is long,and the key sentences of the answers are scattered in multiple paragraphs.The existing model has limitations on the underlying design of long text input.Therefore,a multisegment answer key sentence extraction method is proposed.The method consists of two parts : the sentence score of the extended problem words and the discriminant model.The similarity calculation model of the problem and the sentence is constructed by combining the word form matching and the semantic similarity of the words.A block processing strategy is designed to alleviate the lack of long text input information.The correlation discriminant model between question and text is constructed,and the key sentence extraction of multi-segment answer is realized.Compared with other methods on the data set constructed in this paper,the experimental results show that the performance of this method is better than other comparison methods.(3)Answer key sentence rewriting method based on prompt learningThere is a certain gap between the answers obtained based on the key sentence extraction method of multi-segment answers in prose and the standard answers,specifically,poor generality,excerpting the original text,and inconsistent sentences.It is necessary to rewrite the key sentences extracted from the original text to improve the coherence of the answers.This paper designs a method of rewriting key sentences based on prompt learning.The method consists of two parts: multi-segment answer key sentence extraction module and rewriting module.In order to make the generated text related to the answer and cover the key points of the answer,this paper designs a model of context prompt and continuous reverse prompt strategy guidance.By introducing contextual cues,the amount of information related to the topic can be expanded,so that the answer points can be covered as much as possible when rewriting the answer;by introducing continuous reverse prompts,you can select the most relevant answer to generate a sentence and make the generated answer controllable.(4)Prose short answer question answering systemBased on the multi-span answer key sentence extraction method proposed in this paper and the key sentence rewriting method based on prompt learning,a question answering system for essay short answer questions is designed and constructed.The system has clear business logic,simple and efficient use,simple and beautiful interface.According to the questions raised by the prose,it can be better answered,which can provide auxiliary reference for the problem solving of the college entrance examination prose.