Research On Automatic Answering For Prose Opinion-Question

Machine reading comprehension requires that the machine has the ability to understand and infer text,and to answer corresponding questions based on given background materials.Most of the existing reading comprehension researches focus on corpora in major open domains.The background material description is concise and straightforward,and the questions are clear.The prose background material sentences are subtle and beautiful,with profound meaning,and the questions are more abstract and more challenging.Because this article researches on the automatic answering method of opinion questions in reading comprehension of college entrance examination,the main research work is as follows:(1)Analysis of reading comprehension opinion questions.Through the statistics of the types of reading comprehension questions in Chinese college entrance examinations over the years,it is observed that the scores of opinion questions are relatively high.After further example analysis,it is observed here that the prose genre expression of the background material in the opinion question is implicit and intentional,the question expression is more abstract,strong-generalization,and the question is asked in various ways.After manual classification,it divides them into evidence and reason.In addition,in order to support the machine's research on opinion questions,a corpus of opinion questions is established in this paper.(2)Evidence-type question answering method based on question vocabulary expansion.Because the expression of evidence-type questions is complex and abstract,and the background material is rich and implicit,it leads to a semantic gap between questions and answers.Aiming at this problem,this paper proposes a multi-HLSTM model-based question vocabulary expansion method.First,the background material and the question are interactively noticed,and the two tasks of question prediction and answer prediction are constructed at the same time,so that the modelfurther expands the question.Finally,the extended question and the original question are applied to the answer sentence extraction of the question at the same time.The experimental results show that the problem expansion model can improve the extraction answer sentences' performance of evidence-type question.(3)Cause-type question answering method fusing emotional features.The solution to answer cause-type questions needs to consider the causal logic relationship between the answer and the question.By analyzing the questions and answers in the corpus,it is found that the emotion expressions of these questions and answers are often consistent.Therefore,this paper constructs a method based on pairs of answers and question sentences to help answer cause-type questions.The proposed method extracts the emotional features of question sentences and answer sentences,and integrates them into the embedding layer of the BERT model to help identify the classification of the "question sentence-answer sentence" sentences in a single reading.The experimental results show that the proposed identification method based on the question and answer sentence pairs improves the accuracy of the extracting reason question sentence.(4)Opinion question answering system.Using the evidence-type question answering method based on the expansion of question vocabulary and the cause-type question answering method fusing emotional features,a question answering system for reading comprehension of college entrance examination essay is constructed.The system has a simple interface and clear division of functional modules,which can show the practicability of the method proposed in this paper.