Method for Sentiment Text Analysis based on Statistical and Semantic Properties of Words

In the paper the authors study text sentiment analysis. The research reviewed the existing methods for sentiment analysis. This analysis showed that the most frequently used for this purpose methods of machine learning (such as support vector machines) and classification by linguistic rules. For this reason, the authors decided to combine machine learning based on the statistical properties of the words and linguistic rules. The paper considers a method of sentiment analysis, based on a probabilistic model of the occurrence of words in documents. It proposes to extend this model through consideration of the semantics of words. For the method study, the authors used three collections provided by the organizers ROMIP: reviews of movies, reviews of books from site imhonet.ru, reviews of cameras from Yandeks. Market. The method was further tested on a collection of blog posts provided ROMIP. The research showed that the generalized method based on statistical and semantic properties of the words occurrence in a document provides a fairly accurate assessment of sentiment text analysis.