Optimization of Sentiment Analysis Methods for classifying text comments of bank customers

Abstract

A method of sentiment analysis of the text and its approbation in solving the problem of analysis of text comments left by the Bank's customers are performed. The proposed method consists in a combination of three approaches: rules-based, dictionaries and machine learning with a teacher. New method of text vectorization- tonal vectorization instead of classical ones, such as “bag-of-words ” and TF-IDF, is proposed. The text was classified by logistic regression with regularization. A series of experiments were carried out and the optimal value of the regularization parameter was found in terms of classification accuracy. © 2018