Optimization Of Neural Question Generation Method Based On Seq2seq

With the rapid development and application of natural language processing,the generation of neural problems in the field of natural language processing has also become one of the popular research papers.Neural question generation is to use deep neural networks to extract target answers from a given article or articles,and generate questions based on the target answers.Currently,neurological problems play an important role in learning environments,data mining,information extraction,and countless other application fields.There is a problem in the previous neural problem generation model,that is,the generated question is not clearly related to the substance of the target answer,resulting in a large part of the generated question contains the target answer,which leads to the generation of unexpected questions,and the generated questions are accurate Degree is not high.At the same time,in the absence of guidance,many neural problem generation models do not control the level of specificity of the problems that result,resulting in general and general problems in which information cannot be obtained.In this paper,a neural question generation model based on seq2 seq is used.The model is composed of an encoder and a decoder.We separate the question and the answer,replace the original target answer with a special label,and use the majority and the target answer as input to reduce The number of incorrect questions(including correct answers).Through the copying mechanism based on the overlapping of characters,we can make the generated questions have a higher degree of overlap and relevance at the word level and in the input document.At the same time,we generate a model for the neural classification problem,and train a classifier that annotates a classification problem with a specific level(general or specific)given a certain value.Experiments show that the performance of our proposed model is past,and training a classification problem generation model on special annotated data can generate problems with different specificity levels for a given constant.