Research And Implementation Of Sentiment Summarization Technology For News Field

With the rapid development of science and technology,people get used to use electronic devices to read,make more and more news spread on the Internet,thus leads to information overload problem is becoming more and more serious,at the same time,many emerging from media birth and development constantly,but its content quality is uneven,so monitoring the public opinion information is particularly important.Based on this,the emotion summarization technology in the field of news came into being.Based on the traditional text abstract model which only considers semantic information,this paper takes full account of the emotional information of the text,so as to improve the quality of the abstract and the accuracy of the emotional tendency.It aims to simplify a lot of redundant information in the news field and reduce the reading fatigue of readers.Assist to solve the text creators for the title or subtitle of the creation of difficult problems;At the same time to the website operators for the overall public opinion monitoring to provide help.The main research contents of this paper are as follows:(1)This paper establishes a data set of sentiment summaries in the field of news.The open source data set on the Internet is combined,the data set is compressed according to the conditions of text length and sentence number,and then the sentiment analysis technology is used for sentiment labeling and secondary compression of the data set,and the final sentiment summary Chinese data set is formed.(2)Extractive sentiment summary model is proposed.An extracted text sentiment summarization model was established by combining multi-feature and Lex Rank two-layer graph model.In other words,a Lex Rank two-layer graph based on affective similarity and semantic similarity is established to obtain candidate summaries while considering multiple text features.Then the maximum edge correlation algorithm is used to get the final abstract.The model is trained through the newly established data set,and the corresponding experimental results are obtained,and then the results are compared with other algorithms for verification.(3)Propose generative sentiment summary model.Based on the encoding-decoding model,a generative sentiment summarization model is designed.In the encoder layer,semantic information and emotional information are integrated,attention mechanism is used to solve the problem that the intermediate state vector may lose the early carrying information,pointer generation network is used to solve the problem of unregistered words,and coverage mechanism is used to solve the problem of information repetition,and some improvements are made to this problem.Finally,the new data set is used to train and compare with other algorithms.(4)Establish the emotional summary system in the field of news.An emotion summarization system in news domain is established by using the extracted sentiment summarization model,the generative sentiment summarization model and the sentiment analysis model.Through the design and implementation of text summary generation and text emotional tendency analysis and other functions,including ordinary user page and background management page,to provide diversified services for different audiences.