A Hybrid Sentiment Analysis Method

Sentiment analysis has attracted a wide range of attentions in the last few years. Supervised-based and lexicon-based methods are two mainly sentiment analysis categories. Supervised-based approaches could get excellent performance with sufficient tagged samples, while the acquisition of sufficient tagged samples is difficult to implement in some cases. Lexicon-based method can be easily applied to variety domains but excellent quality lexicon is needed, otherwise it will get unsatisfactory performance. In this paper, a hybrid supervised review sentiment analysis method which takes advantage of both of the two categories methods is proposed. In training phrase, lexicon-based method is used to learn confidence parameters which used to determine classifier selection from a small-scale labeled dataset. Then training set which is used to train a Naive Bayes sentiment classifier. Finally, a sentiment analysis framework consist of the lexicon-based sentiment polarity classifier and the learned Naive Bayes classifier is constructed. The optimal hybrid classifier is obtained by obtaining the optimal threshold value. Experiments are conducted on four review datasets.