Sentiment Orientation Analysis Of Chinese Sentence

In recent years, information resources on internet expand rapidly. There are more and more remarks with personal feeling, they are in the form of BBS, blog and so on. Contents of these remarks are mostly comments to products, movies, characters and so on, which express their subjective views of things. Analysis of these texts is of great practical significance, so sentiment orientation analysis of text has become a research hot spot of current internet intelligent information processing. It involves natural language processing, information retrieval, information extraction, machine learning and so on.Sentiment orientation analysis is to determine its sentiment polarity and degree. The analisys result can be devided into three classes:positive, negative and neutral, or five classes by sentiment intensity. In this paper, I focus on the internet review sentence, including sentiment word and phrase in the review.The main content of this paper can be listed as following. Firstly, for the sentiment word analysis, I construct sentiment word dictionary, introduce a synonym-based SO_PMI algorithm and HowNet approach. Secondly, for phrase sentiment analysis, I extract the adverb and target word related sentiment word by some rules(for example, dependency relationship); and then get sentiment value of phrase according to the sentiment word. At last, it comes to sentiment analysis for sentence. On the one hand, I calculate it through the sum of weighted sentiment values of all phrase. On the other hand, traditional text classification mathod is used with some improvement in feture selection and disposal after the corpus is classfied.The results shows that precision and recall rates are improved a lot both in word, phrase and sentence.