Sentiment Classification into Three Classes Applying Multinomial Bayes Algorithm, N-Grams, and Thesaurus

The paper is devoted to development of the method that classifies texts in English and Russian by sentiments into positive, negative, and neutral. The proposed method is based on the Multinomial Naive Bayes classifier with additional n-grams application. The classifier is trained either on three classes, or on two contrasting classes with a threshold to separate neutral texts. Experiments with texts on various topics showed significant improvement of classification quality for reviews from a particular domain. Besides, the analysis of thesaurus relationships application to sentiment classification into three classes was done, however it did not show significant improvement of the classification results.