Abstract: Cross-domain sentiment classification has attracted more attention in natural language processing field.Given that tradition active learning can’t make use of the public information between domains and the bag of words model can’t filter these words not related with sentiment classification,a method of cross-domain sentiment classification based on optimizing classification model progressively was proposed.Firstly,this paper selected the public sentiment words as features to train classification model on the labeled source domain,then used the classification model to predict the initial category label for target domain and selected the texts with high confidence value as initial seed texts of the learning model.Secondly,we added the high confidence text and low confidence text to the training set at each iteration.Finally,the feature set was extracted to transform feature space based on the sentimental dictionary,evaluation collocation rules and assist feature words.The experimental results indicate that this method can not only improve the accuracy of cross domain sentiment classification effectively,but also reduce the manual annotation price to some extent.

Key words: Sentiment classification,Cross domain,Classification model,Feature extraction,Confidence