The Research On The Classification Of Questions Based On Convolution Neural Network

The realization of natural language understanding and even language generation has always been one of the ultimate goals of the linguistic and computer science.Natural Language Processing(NLP),both in computational linguistics,has been exploring the internal coding of the language and trying to decode it.Question classification is an important part of natural language understanding or natural language generation.Its classification system affects the whole question analysis module,and the accuracy of question classification will directly affect the subsequent module of language understanding.Starting from this point,this paper will combine the hierarchical structure of question classification system and convolutional neural network model,some factors according to the experimental data,the specific system of multi-level question classification rationality and influence the classification accuracy of detailed analysis and demonstration.The first chapter is the introduction,first introduced the composition of question answering system,question classification module,in which has the position;second describes the research background,research status of the question answering system and question classification respectively at home and abroad;then this paper expounds the research significance from three aspects;finally the research organization in this paper,the structure of A.The second chapter is divided into three parts.The first part is mainly the process of selecting the data,including data selection criteria and selection method;the second part is the analysis of the level of interrogative sentence,interrogativesentence to classify from the traditional linguistic perspective,and establish amulti-level classification system--M-QCS;the third part is the data preprocessing,including stop words,word segmentation words,vector conversion.The third chapter is divided into three parts.The first part introduces the convolutional neural network(CNN)some of the theoretical background,including deep learning(Deep Learning)the basic theory and the development of CNN;the second part is mainly introduce the basic architecture of CNN,including the convolution layer,activation function,pooling layer;the third part is to construct CNN with a general description of the idea of hierarchical design model.The fourth chapter is divided into two parts.The first part is mainly divided into three aspects,one is to introduce the artificial selection corpus standards,including questions of different types of constructions;the two is to show the training model of Word2vec,and briefly describe the principle of CNN model specific;three is the experiment of some process,including some parameters and some pseudo code.The second part is the data statistics of the experimental results and the analysis of the experimental results.First,the rationality of the multi-level classification system is verified by the experimental results,and the reasons for the classification errors are analyzed.Secondly,through word frequency statistics for data analysis,it is concluded that the frequency of high-frequency words is positively related to the accuracy of classification,and the reasons for classification errors are also analyzed.The fifth chapter is roughly the overview of the views of this article,combing the full text of the context,and further summary of the conclusions of this article.Including some views about the inadequacies of this article,mainly from the perspective of linguistics and computer science to explain;the second is for some minor linguistic problems involved in this paper to sort out,provides some ideas and directions for future research on question classification.