Abstract: According to the structure, the news style was divided into several paragraphs. Based on sentiment lexicon and semantic rules, a method of extracting sentimental key sentences was used to analyze the sentiment of sentences within each paragraph. Firstly, sentiment lexicon was built by considering the emotion, twist, negation, degree and sums up vocabularies; Secondly, according to rules, news text was divided into sense groups, sentences, paragraphs and chapters; Furthermore, orientation value of sentimental key sentences was computed by the rules established, and then the sentimental orientation value of the paragraphs and the whole chapters was obtained by weighted average of sentences, thus the sentimental orientation of news was revealed. Compared with lexicon based method and SVM sentiment classification, experimental results show that the method proposed has good effects on the orientation identification of news text, showing good feasibility as well.

Key words: rules, sentiment lexicon, sentiment analysis, online news