Research On Automatic Abstraction Method Of Single Document Based On Hybrid Neural Network

In recent years,with the development of network,massive data sets and electronic documents have appeared on the Internet,and people are increasingly relying on Internet to obtain information.Therefore,the field of automatic summarization has been greatly expanded,and the use of summarization is also more and more extensive.The traditional summarizations are all formed by the artificial understanding in target text,which takes a great deal of time and effort,and has heavy workload,which can no longer meet the requirements in information age.In response to the problem,automatic summarization has been produced.Recently,the application of deep learning methods in automatic summarization has gradually become a new research hotspot.How to quickly and effectively obtain practical information from the massive information library has become an urgent problem,and automatic summarization is one of the powerful tools to solve this problem.To solve the problem of text content representation and abstract content selection,an automatic summarization method based on hybrid neural network model is proposed.Aiming at the sentence extraction in automatic summarization,an improved automatic retrieving method based on recurrent neural network is proposed to improve the performance of the model.The research work of this paper mainly includes the following four aspects:(1)Because the improvement space is limited in text preprocessing technology,and the text content representation and abstract content selection can be improved to a large extent,an automatic summarization method based on hybrid neural network model is proposed in the paper.The proposed method combined convolutional neural network with long-and short-term memory network model,the former(convolutional neural network)is efficient and difficult to overfit during training,and the latter(long-and short-term memory network)has good effect on sequence prediction,where,the former represents sentence vector and the latter is used to extract the summary sentence.(2)In view of the sentence extraction in automatic summarization process,this paper adopts an improved cyclic neural network model,which replaces the hidden layer in original recurrent neural network with the LSTM memory cell structure.The experimental results show that the obtained ROUGE-2 value and ROUGE-3 value by proposed model is higher 0.024 and 0.0155 respectively than improved recurrent neural network based on LSTM.(3)The performance of six models are compared,they are LDA model,LSI model,decision tree,logistic regression,neural network,convolutional neural network and improved LSTM cyclic neural network based on automatic abstraction generation,respectively.The experimental results show that the single document automatic summarization method based on hybrid neural network works best.(4)This paper designs and implements a single document automatic summarization system based on hybrid neural network.The system is divided into three parts.The first part is the word segmentation processing in the text data.The second part is the word turn vector of the result after word segmentation.The last part is to call the automatic summarization algorithm of this article to get the text summary after calculation.