Research On Text Summarization Technology Based On Deep Learning

The fast construction of mobile communication and Internet facilities in China promotes the rapid popularization of mobile Internet based on mobile communication and Internet technology,then has brought the prosperity of social media applications operating on mobile Internet.The prosperity of social media applications has reduced the threshold of Internet applications,and more people have become the creation center of Internet information,enriching the content of the Internet.However,the full and colorful content on the Internet has brought obstacles to people's efficient access to information under the rapid pace of production and life.High quality text summarization can improve the efficiency of information selection and acquisition.Since the enhancement of computer hardware performance and the development of deep learning,the application of deep learning to automatic text summarization becomes more and more important.At present,automatic text summarization can be divided into extraction and abstraction,according to the way of generating summarization.The summarization,obtained by extractive method,are sentences in the original text,which are different in length and not strong in generality.Abstractive text summarization method is similar to people's understanding of the original information.The length of sentences can be controlled,but it is difficult to fully capture the main information of the original.To solve these problems,this paper uses deep learning to further study the abstractive text summarization method.The main details in research are as follows:(1)This paper makes a further research on the sequence to sequence model,including the structure,principle and types of the sequence to sequence model.And the paper points out the shortcomings of the sequence to sequence model in some cases where it is difficult to obtain an accurate semantic representation of the summary.Aiming at this problem,the attention mechanism is introduced to improve.(2)This paper makes a deep research on the mechanism of attention machine,including its principle and essence,and further studies the basic principle of self attention mechanism.On this basis,we propose to improve the sequence to sequence model by using a hybrid attention mechanism which can take into account global attention information and local attention information,then get a text summarization model combining the above two.During the training,the method of adversarial learning is used to dynamically adjust the supervision strength of the model to further improve the effect of the model.(3)After a brief introduction of the data set,this paper analyzes the related data set and selects the appropriate data set.Based on the proposed model,experiments are designed on selected data sets,and the evaluation method rouge,which is commonly used in text summarization.The results are compared with baseline to prove the effectiveness of the model.