Research And Implementation Of Text Automatic Summarization Based On Deep Learning

In the fast-food era of online culture,data resources are growing explosively,which brings people the trouble of information overload.With the acceleration of social rhythm and the increase of life pressure,people do not have enough time and energy to browse all text information,and often want to use Fragmented time to get more useful information.By reading the text summary,people can quickly grasp the main idea of the article,determine the necessity of the article,and will not be deceived by the article title party,and effectively alleviating the time and energy problems caused by information overload.The main purpose of the text automatic summary is to convert a text or collection of text into a short summary containing key information,divided into categories of s,which can be divided into indicative summaries,information summaries,keyword summaries,and title summaries.With the development of deep learning in the field of natural language processing in recent years,automatic summary research based on the Seq2Seq framework has become mainstream.Therefore,based on the Seq2Seq framework,this paper introduces attention mechanism and constructs a keyword summary model and an information summary model.The main research contents are as follows:(1)Keyword summarization research based on deep learning.The text semantic representation based on word2vec word vector and the keyword abstract method based on Seq2Seq framework are studied,and introduces the attention mechanism as the baseline model for experiments,compared with the traditional machine learning method,and on this basis,introduces the copying mechanism,which copies the appropriate segments from the input sequence to the output sequence.The experimental results show that the copy mechanism based on the baseline model can improve the effect of keyword summary generation.(2)Research on information summarization based on deep learning.The information abstract model DVSNET based on semantic similarity calculation is proposed,which mainly focuses on the problem that the baseline model can not decode unregistered words and the common repetitive problems of the generated tasks,the pointer network model and coverage mechanism are introduced,and in order to further improve the quality of automatic summary,document similarity calculation is added to maximize the semantic similarity between the target summary and the original text.The experimental results show that the model can solve the non-login problem in the message summary and alleviate the problem of repetition and smoothness in the task of generative summary.(3)Design and implementation of automatic summarization system.This paper introduces the overall design ideas and processing flow of the judicial case intelligent retrieval system,and applies the generated keyword summary to the system,builds a natural language processing platform,realizes the front-end page system display of the information summary,and proves the practicability and effectiveness of the automatic summary.