?
Alternative method sentiment analysis using emojis and emoticons
Our research aims to develop an alternative method for analyzing the tonality of the texts. Most of the traditional methods for determining tonality classes are based on text analysis and ignore various emotional indicators that users actively used in social networks. Therefore, it improves the quality of predicting the tonality of the class. The study is focused on three methods of expressing emotions in a text, emojis, emoticons, and punctuation marks that express emotions. We have developed a special lemmatizer for data preprocessing and built several text classifications models to classify the text into two classes, positive and negative, where emotional indicators are used as predictors. We have also used the RuSentiment corpus to create the classifier. The study has demonstrated that the proposed methods improve prediction of tonality classes by 6% compared to the traditional models. We have obtained the best results using a model ensemble based on the emotional indicators model and the Word2vec model. The model has demonstrated convincing results 91% accuracy and 0.937 area under the ROC curve. We have identified several patterns. Emotional indicators have a pronounced connection with tonality classes. Positive emotions are much more than neutral and negative. The model of emotional indicators has 85% accuracy; therefore, emotional indicators are very crucial in the analysis of the text tonality.