News-text Sentiment Classification Based On Jst Model

With the rapid development of Internet, network brings together a lot of information, including network news. Network news has become an important information carrier and a very important channel for the access to information. Due to network news content rich and varied and sentiment tendencies also are different, the automatic analysis to the network news content has become a hot region in the field of text processing in recent years.At present, most materials of the sentiment analysis are from the comments. Sentiment tendency of comments are rather obvious and comments are mostly short text. Relative to the comment text, the news text is a description of an event with weaker subjective, and mostly are long text. In view of the above characteristics, this paper made a study on the sentiment of news analysis, the main work is as follows:We analyze the sentiment for news based on JST model. We verify the feasibility of JST model in the news text. The JST model is unsupervised and needn’t to mark the training sample. Therefore it does not exist the problem of the field transfer. To join prior knowledge-appraise dictionary, JST model can further improve the classification accuracy.The news title is the summary of the content of the news, sometimes with certain sentiment tendencies. Therefore we add the news title polarity analysis into news sentiment analysis. At first we find polarity of news title, and then combine the title polarity with JST model to make sentiment classification. There are two methods for title polarity: sentiment dictionary and word semantic.News is usually long text. There are a lot of irrelevant sentiment topic sentences in the news. They will affect the accuracy of classification. This paper presents a method of extracting sentiment topic sentences. First of all, we find the subjective clues applied to the news and give score for subjective sentence; secondly, we give score for topic sentence; finally, considering the sentence score of subjective and topic, we extract of the highest score in the first k sentences as sentiment topic sentences of the document. Instead of the full text of the news we use sentiment topic sentence analysis.The experiment result shows that the method proposed in this paper is effective.