A comparative study on Chinese sentiment classification

Sentiment classification task can be solved in the following two ways: one is based on a supervised learning manner; the other is unsupervised learning approach. The article conducts the related technologies of these two manners for a comprehensive comparative analysis. For supervised learning manner, different preprocessing types, feature selection methods, combined with SVM and KNN algorithm were investigated. For unsupervised learning manner, the influence of different preprocessing type, the selection of reference words, and other factors were analysis. Comparative experimental results with Chinese sentiment classification benchmark ChnSentiCorp show that supervised learning manner efficient than unsupervised manner, but unsupervised learning way have more stable performance.