Program of Texts Sentiment Analysis Based on Syntactic Dependencies Trees with Machine Learning Methods

The purpose of this paper is to research the binary sentiment classification of Russian texts using syntactic features. Base approaches to sentiment classification and to syntactic dependencies extraction are analyzed in this work. In addition, variety of methods to extract features for sentiment classification are provided. The aim of the work was to develop a program of texts sentiment analysis based on syntactic dependencies trees with machine learning methods. Methods of texts pretreatment and approaches to the representation of the text information in vector form are implemented. Approach to syntactic features extraction is implemented, also the naive Bayes classifier and K-nearest neighbours algorithm are described. In order to assess the quality of the algorithms key metrics are described and implemented. According to the results assess of the quality of algorithms with different sets of features for classification, syntactic features provided a significant improvement in the quality of the classification. The work have passed approbation at two conferences.  The paper contains 66 pages, 3 chapters, 5 illustrations, 35 tables, 50 bibliography items, 5 appendices. Keywords: natural language processing, sentiment analysis, POS tagging, morphological disambiguation, syntactic analysis.