Semantic Based Sentiment Classification On Blog Community

BLOG is an information issuing format emerged in late 1990s. It was introduced into China in 2002. Within just several years, BLOG has evolved into one of the most active virtual communities growing with high speed. Blogspace, a platform on which millions of people exchange their personal opinions and other information, contains a great deal of precious commercial values. In the field of products investigation, the most import thing is to gain the reviews on the specific product from actual users. Those comments are critical for both manufacturers and latent clients. The core technology to analyze the BLOG posts collected is called Sentiment Classification.Sentiment Classification has vast applying fields, such as analysis of commercial products reviews, public sentiment analyzing and spam filtering. There are two categories of Sentiment Classification methods; one is based on semantics, while the other relies on machine learning methods. Both of the two categories have a drawback that regards the documents as sets of words or patterns. No context information has been saved. So, when the author of the comment cites some arguments of the rivals in the text, those cited arguments will be recognized as the opinions of the author himself, and then wrong judgment made.In the thesis, we design and implement a Sentiment Classification method based on shallow semantics. The method introduces transitional syntax structures recognizing into the sentiment classifying process. First, the transitional complex sentences are figured out and the sentimental values attached on them are adjusted. Then, the transitional sentence groups are recognized, once again, adjustments are made on the sentimental values of transitional sentence groups. Our experiment shows the new method can improve the accuracy of the Sentiment Classification on critical texts on the whole.