Sentiment classification using TF-1DF features and extended space forest ensemble

With the rapid development of electronic commerce, user-generated contents have become increasingly important for customers and suppliers who want to get more feedback. They also have attracted a great deal of attention in the academics. Particularly, making sentiment classifications for these contents is significant. Till now, it has been proved that ensemble method is available for sentiment classification in the theory and practice. Following this direction we propose a new feature construction method by taking advantage of TF-IDF method, and an extended space forest ensemble method under the framework of bagging is employed for sentiment classification. In the experiment part, we make a performance comparison among the extended ensemble method with different feature operators and the original one based on two base classifiers by using public sentiment dataset. The empirical results show that the extended space forest ensemble method with appropriate feature operator can greatly improved the effectiveness of sentiment classification.