RS-LSSVM: A new ensemble method for sentiment classification based on random subspace

As analyzing and predicting the polarity of the sentiment plays an important role in understanding social phenomena and general society trends, sentiment classification problem has become a popular topic in academia and industry in recent years. However, comparing with Bagging and Boosting, another popular ensemble method, i.e., Random Subspace, is paid much less attention to the sentiment classification problem. In this research, we propose a new ensemble method, RS-LSSVM, for sentiment classification based on Random Subspace and LSSVM. Ten public sentiment classification datasets are used to verify the effectiveness of the proposed RS-LSSVM. Experimental results reveal that RS-LSSVM can get the better results than the four base learners, Bagging, and Boosting. All these results indicate that RS-LSSVM can be used as an alternative method for sentiment classification.