Three language political leaning text classification using natural language processing methods

In this article, the problem of political leaning classificationof text resource is solved. First, a detailed analysis ten stud-ies on work’s topicwas performed in form comparative characteristicsof used methodologies.Literary sources were compared according to problem-solvingmethods,the learning that was carried out, evaluation metrics, and vectorizations.Thus, it determined machine algorithms neural networks, as well vectorizationmethods TF-IDF Word2Vec, most often solve problem.Next, various classification models whether textual information pro-Ukrainian or pro-Russian built based dataset containing messages from social media users about events large-scale Russian invasion Ukraine February 24, 2022.The solved with help Support Vector Machines, Decision Tree, Random Forest, Naïve Bayes classifier,eXtreme Gradient BoostingandLogistic Regressionmachine algo-rithms, Convolutional Neural Networks, Long short-term memory BERT techniques for working unbal-anced dataRandom Oversampling, Undersampling , SMOTE SMOTETomek, stacking ensembles models.Amongthe algorithms, LR best, showing macro F1-scorevalue 0.7966 when features trans-formed by vectorization 0.7933 BoW.Among best 0.76was ob-tained using CNN LSTM.Applying data balancing failed improve results algorithms.Next, determined. Two constructed achieved same LR. Ensembles wasable do so consisted vectorization, B-NBC meta-model, SVC, NuSVC LR, base models, respectively.Thus, three classifiers, algorithmand two which defined combination existing methods solving problem, demon-strated largest F1-score value 0.7966. The obtained can be review news publica-tions characteristic, people identify being isolated filter bubble.