New algorithm for determining the number of features for the effective sentiment-classification of text documents

Sentiment analysis of text documents is a very important part contemporary mining. The purpose this article to present new technique sentiment which can be used with any type document-sentiment-classification method. proposed involves feature selection independently classifier, reduces the size space. Its advantages include intuitiveness and computational noncomplexity. most element novel algorithm for determination number features selected sufficient effective classification. based on correlation between single document labels. A statistical approach, featuring naive Bayes classifier logistic regression, was employed verify usefulness technique. They were applied three sets composed 1,169 opinions bank clients, obtained in 2020 from Poland-based bank. written Polish. research demonstrated that reducing terms over 10-fold by means cases improves effectiveness