Research And Implementation Of Automatic Generation Method Of Internet News Abstract

With the development of an intelligent society,the way people come into contact with various news in their daily lives has gradually changed from traditional media such as newspapers and magazines to various smart terminal media.Among them,various application software on smart phones is one of the important tools for people to obtain information.However,for people who need to obtain news information in a specific field,only part of the Internet news content meets their needs.The summary of news can help people achieve the purpose of quickly screening news content,so as to reduce the time cost for people to obtain news.The accuracy of the news content text will have a direct impact on the quality of the abstract.This thesis expands and explores the problem based on the existing theory.The work done is as follows:Firstly,it has studied in depth the proofreading technology of Chinese text and applied it to the preprocessing process of text summary generation.This thesis studied comprehensively the text proofing technology of word errors and semantic errors,and used the Spell GCN spell check method to find word errors in Chinese texts.After that,we extracted multiple features from a large amount of training corpus,and trained the support vector machine(SVM)classifier to correct word errors.Uni LM model technology is used to correct the semantic errors of Chinese text,and reconstruct the sentence through text generation based on extracting the semantics of the sentence.The experiment verifies the effectiveness of the Chinese text proofreading method.Secondly,this thesis studied the application of deep learning models in the automatic generation of Internet news summaries.A filtering method based on N-gram model and semantic vector is designed,and the output is expanded by beam search algorithm to generate a candidate set.The best candidate from the candidate set is selected as the output according to the designed filtering algorithm.The experimental results show that the improved strategy can significantly improve the performance of the model results.Thirdly,this thesis designed and implemented an automatic abstract generation system for Internet news whih consists of a client and a server.The client is an application carried on an Android smart phone,which realizes the function of generating a corresponding summary of the news content input by the user.The server side is the neural network model and the corresponding functional interface deployed on the server.This thesis presents the key parts of the various algorithm processes and system implementation used,and verifies the effectiveness of the algorithm and the usability of the system through experiments.