Research And Application Of Automatic Text Summarization Technology Based On Deep Learning

Under the influence of the reform and opening-up policy over the past four decades,China’s society has developed rapidly.In particular,in the 1990 s,China introduced Internet technology,which further enriched people’s social and cultural life.People can get all kinds of information by surfing the Internet.However,the Internet produces a huge amount of information data every day,and only a very small part of this large amount of data is the information we need.How to filter redundant and invalid information from these massive data,and obtain useful information for oneself quickly and accurately,is a matter that requires a lot of cost.People are also looking for ways to save the cost of obtaining information.Through the joint efforts of a large number of scientists and researchers,the field of natural language processing has developed rapidly,and automatic text summarization technology has gradually entered the public’s field of vision.Text summarization technology analyzes and studies the structure and characteristics of the text,so as to extract the central sentences of several texts to form the text summary,or learn and train the lexical,grammar,and semantics of the text to understand the expression and meaning of the original text.,which in turn generates sentence summaries that generalize the original text.Starting from the service objects of the project,this research proposes to use text summarization technology to process information such as news and articles involved in the project according to the characteristics of the elderly.Corresponding text summaries are generated from these long text information,thereby improving the speed at which the service object obtains information and saving various costs of obtaining information.During the research process of this paper,I learned some classic ideas and ideas of predecessors,deeply studied the execution process and general idea of some classic models,and compared the advantages and disadvantages of each model.For natural language processing tasks,it is generally divided into upstream word vector processing in the first stage and downstream task-specific fine-tuning in the second stage.The main innovation points of this paper are as follows: First,in the process of word vector processing,the effect of the model is improved by improving the mask strategy of the classic model.Second,according to the characteristics of the corpus environment used by the project,the processing of word vectors is replaced with another model,so that the generative pre-training model originally used in the English environment can handle the task of the Chinese corpus environment.Third,by fine-tuning and improving existing generative pre-trained models,the models can handle text data of different lengths.Finally,a suitable Chinese data set was selected for a series of experimental verifications,which achieved good results in the processing of short text and medium and long text data,but there is still a lot of room for improvement in the processing effect of super long text data.Overall,the use of the improved model can meet the functional requirements of this study.