Research On Problem Generation Technology Based On Deep Learning

Question generation is an emerging research direction in the field of natural language processing.The research aims to enable machines to read and understand human language and automatically generate high quality questions to help people improve their information needs.In real life,the application scenarios of question generation technology are very extensive.For example,in smart customer service,smart medical,knowledge quiz games,etc.,question generation technology has great application prospects,which can help people to liberate from complicated work.Improve people’s quality of life and work efficiency.As a popular technology in recent years,deep learning has achieved great success in image,video and natural language processing.However,in the task of question generation,the application of deep learning is still less.This paper attempts to apply deep learning technology to question generation,and carries out exploratory research.Two kinds of question generation frameworks and a question generation system based on deep learning are proposed.The detailed work is as follows:(1)An end-to-end question generation model based on encoder-decoder is proposed(Question Generation based encoder-decoder model,EDM framework for short).The traditional rule-based question generation method relies heavily on the feature set captured manually.The generated questions usually have some defects,such as incompatible sentences,unreasonable semantics,word repetition,and mismatch with the content of the article.The method is not universal enough.In view of the above situation,this paper proposes an end-toend question generation model based on encoder-decoder.The model includes the structure of encoder-decoder based on attention mechanism,Pointer-Generator Network,etc.,And for the particularity of the task of question generation,a series of designs for the network structure.which can be well applied to question generation tasks.(2)An improved question generation framework is proposed.The input of EDM framework is articles,which only generate questions by articles,but not make full use of answer information,which may lead to low quality of generated questions.Therefore,on the basis of the EDM framework,we add a part that can handle the answer information separately,and explicitly input the answer as a separate part to the model,and make a series of improvements to the network structure.From the experimental results,the improved question generation framework is better than the EDM framework.Because the input of the improved framework is articles and answers,it is called the Answer Aimed Question Generation Model(AAM framework).(3)Based on AAM framework,a question generation system is developed.The system includes data input visualization module and background data processing and decoding module.