Machine learning in sentiment-analysis of text information on the example of user attitudes regarding candidates for ukrainian presidential elections 2019

Abstract

The main methods of machine learning for the sentiment analysis of the text are described and a comparative analysis of their effectiveness is performed. The stages of pre-processing of the text, such as stemming, deletion of stop words, algorithms for converting the text to vector form, such as bag-of-words (Bag-of-Words), TF-IDF vectorizer and Word2Vec, are considered. The goal of this study was to determine the sentiment of the comments under the publications of Ukrainian Presidential candidates (V. Zelensky and P. Poroshenko) during the 2019 election campaign.Three algorithms were used to determine the tonality of the text: the naive Bayes classifier, the support vector machine, and the convolutional neural network. Separate models were built for each candidate and a comparison of the classification quality was performed (according to metric F1). The most precise model for both data samples was a convolutional neural network.