Research On Question Similarity Of Intelligent Question Answering System Based On Deep Text Matching Model

In the rapidly developing Internet era,with the continuous improvement of people's living standards,people's needs in the field of artificial intelligence such as information retrieval,automatic question answering,and dialogue systems are getting higher and higher.Intelligent matching algorithms are needed to meet the diverse needs of users.To solve this problem,natural language processing technology came into being,which can provide users with efficient information retrieval services and comfortable human-computer interaction experience.Text matching task is the core problem in natural language processing technology.In recent years,with the rapid development of deep learning and text word vector technology,text matching based on deep neural networks has gradually become a new research direction.This paper studies some classic deep learning models for text matching.Based on the four models of ESIM,DSSM,Decomposable Attention,and Siamese Network,a fusion model is proposed based on the Blending model fusion method and adding a layer of logistic regression on the Blending last.The main tasks include:First,the related technologies of text matching are studied,including word vector technology,convolutional neural network,recurrent neural network,two-way recurrent neural network,and model fusion technology.Secondly,the basic structures of the four models are introduced respectively,and on this basis,the model construction is carried out for the actual problems in this paper,the network layers of the four models are explained in detail,and the loss function and optimization algorithm are explained.Explain the construction process of the model proposed in this paper.Finally,experiments are performed based on the 2018 ATEC NLP competition data.The experimental results show that the ESIM model performs best in actual text matching tasks.In the final model fusion phase,based on the Blending model fusion method,logistic regression was used as the last layer of Blending for model fusion.The model was applied to the test set with an accuracy rate of 0.716537,a recall rate of 0.748,and a of 0.725961,all higher than four Base model.The experimental results show that the model designed in this paper can use the Blending model fusion method to make full use of the differences between network architectures,extract the information in the text,give play to the advantages of each model,and improve the accuracy of the model.