Abstract: Currently, a huge number of texts are being generated, and there is an urgent need to organize them in a certain structure in order to perform classification and correctly define categories. The authors consider in detail such aspects of the topic as the classification of texts in natural language and the definition of the tonality of the text in the social network Twitter. The use of social networks, in addition to numerous advantages, also carries a negative character, namely, users face numerous cyber threats, such as personal data leakage, cyberbullying, spam, fake news. The main task of the analysis of the tonality of the text is to determine the emotional fullness and coloring, which will reveal the negatively colored tonality of speech. Emotional coloring or mood are purely individual traits and thus carry potential as identification tools. The main purpose of natural language text classification is to extract information from the text and use processes such as search, classification using machine learning methods. The authors separately selected and compared the following models: logistic regression, multilayer perceptron, random forest, naive Bayesian method, K-nearest neighbor method, decision tree and stochastic gradient descent. Then we tested and analyzed these methods with each other. The experimental conclusion shows that the use of TF-IDF scoring for text vectorization does not always improve the quality of the model, or it does it for individual metrics, as a result of which the indicator of the remaining metrics for a particular model decreases. The best method to accomplish the purpose of the work is Stochastic gradient descent.