Abstract
Sentiment classification is an important data mining task. Previous researches tried various machine learning techniques while didn't make fully use of the difference among features. This paper proposes a novel method for improving sentiment classification by fully exploring the different contribution of features. The method consists of two parts. First, we highlight sentimental features by increasing their weight. Second, we use bagging to construct multiple classifiers on different feature spaces and combine them into an aggregating classifier. Extensive experiments show that the method can evidently improve the performance of sentiment classification.