Sentiment Analysis On News Text Based On Key Sentiment Sentences

With the popularity and rapid development of the Internet, the network has become a major kind of modern information carrier, through analysis of online text, especially the news text online, we can clearly grasp current opinion trends, mine public interests and filter the information etc. The text sentiment polarity analysis has risen with the arrival of Web2.0era whose contents are subjective and generated by users. However, current studies didn’t take the sentimental features and distribution into consideration, so document-level sentimental classification accuracy is difficult to achieve the accuracy of common text classification.In this paper, based on existing research, we put forward a method using sentiment polarity, keywords, positions and similarity between sentences and titles to classify sentiment with the help of study for news. In this way, we compress the feature space of document. Here are what we do: First, extract key sentiment sentences:Compared with other text, news whose theme is very clear has its own unique features. In this paper we proposed a method to describe key sentiment sentences, as well as how to extract every kind of sentimental features.Second, document-level sentiment analysis with compressed feature space:with the help of key sentiment sentences, we can greatly reduce feature dimensions. The variety of machine learning methods are applied to build sentimental classifiers. This issue use BP neural network model and SVM model who can effectively utilize the extracted features on sentimental analysis. This study has practical value.