Abstract: Micro-blog is a new social media platform based on Web 2. 0. Internet users express their feelings, emotions,favorites and disgust through micro-blogs, resulting in a large number of emotional text information. We can know the emotional state of the Internet users, the point of a social phenomenon and preference of a product, through analysis of the emotional text information, which not only has a certain kind of commercial value, and is helpful to the stability of the society. In this paper,we use the emoticons form micro-blogs,combined with emotional words to build the Chinese emotional corpus,ensuring the scale and accuracy of the corpus,eliminating the need for artificial burden. Based on the corpus,we construct Baycs classifier and use the entropy to improve the performance. We compare different performance while changing the type of n-gram. Finally, we get the best classification results using unigrams as features and optimizing with entropy. Recall rate and accuracy can be achieved above 85 0 o，the F measure can even reach more than 89%.

Key words: Sentiment analysis, Emoticons, Emotional words, Micro-blog