Abstract: This paper presents a machine learning algorithm which combines the semantic and statistical aspect fortext sentiment classification.First,extract the text of the emotional wordsas features,and use the statistical methods to get the initial weight,then re-weight the feature by analyzing the semantic structure of the text;finally use Bayesian algorithms and Bayesian classification algorithm as a basic classification of the Boosting algorithm to classify the text.The empirical resultsshow that,the accuracy of the Bayesian classification algorithm based on semantic understanding is higher than the Bayesian classification algorithm based on statistic and its accuracy can reach 90%.

Key words: semantic orientation, Bayesian, Boosting, sentimentclassification