Research Of News Oriented Text Categorization And Sentiment Analysis Technology

With the rapid development of the Internet,more and more Internet media have begun to replace the traditional print media and television media.Users face various kinds of news every day.The classification of the news can help the users to find their own concerns from the mass of data.The sentiment analysis of news helps users quickly understand the situation of an industry or event,so as to assist the user to make some decisions.It is a time-consuming task to classify and annotate the news artificially.Therefore,how to realize the automatic classification and sentiment analysis of the news text is a very valuable research work.The work includes two aspects:1.This thesis proposes Vote-S VM,a news text classification technology based on sentence category voting which combines Word2vec with SVM.Compared with the traditional method,Vote-SVM analyzes the text from the granularity of the sentence and determines the category of the news based on the categories of all the sentences,it has a better classification effect on news classification.This thesis also analyzes the defects of traditional TF-IDF algorithm in extracting category feature words,and improves the traditional TF-IDF algorithm,so that it can better extract the category feature words.The experimental results show that the feature extraction algorithm based on improved TF-IDF can effectively improve the quality of class feature extraction.2.Aiming the lack of completeness of existing sentiment dictionaries,This thesis proposes a sentiment dictionary construction method based on word similarity and SOPMI algorithm,and build a category sentiment dictionary for each news category through this method.This thesis also proposes Senti-SVM,a sentiment analysis method based on category sentiment dictionary and SVM.This method extracts the fine-grained sentiment features of the news according to different categories of news,and can extract the sentiment information in the news more accurately.From the experimental results,the sentiment dictionary construction method proposed in this paper can effectively construct sentiment dictionaries for corpus data,and finegrained sentiment feature extraction based on categories has a positive impact on the improvement of sentiment classification performance.