Multi-view ensemble learning method for microblog sentiment classification

With the rise of microblog services in recent years, microblog sentiment classification has been widely studied and applied in many fields like public opinion monitoring, commodity evaluation and market forecasting. Ensemble methods have been widely used in the feature construction and classification stages of microblog sentiment classification due to their excellent performance. In feature construction, most researchers use feature concatenation or ensemble methods to combine different features while the fusion of two methods is ignored. In classification, most ensemble classification methods combine classifiers based on majority voting or weighted averaging, and they do not fully consider the differences in the information contained in classifiers. In this paper, a novel multi-view ensemble learning method is proposed to fuse the information contained in different features for better microblog sentiment classification. This method consists of two stages: the local fusion stage and the global fusion stage. In the local fusion stage, the raw features and concatenation features are used to construct basic classifiers, and these basic classifiers are combined into five classifier groups to identify the microblog sentiment among all raw feature information. In the global fusion stage, these classifier groups with a global view are further integrated to make more accurate and comprehensive predictions. Two public microblog benchmark datasets provided by Sina Weibo are used in the experiment, and the experimental results show that our method outperforms other compared methods in identifying the polarities of microblog posts.