DETECTION AND CLASSIFICATION OF PROPAGANDA TECHNIQUES AND OBJECTS IN TEXT MESSAGES USING MACHINE LEARNING

Abstract

Article proposes approach to detecting and classifying propaganda techniques and objects in text messages using machine learning models, which consists of performing sequential steps and allows both to detect the use of techniques in general and to classify the applied techniques, as well as to detect objects to which the classified propaganda techniques are directed. To detect the applied propaganda techniques, hybrid machine learning model was used based on the combination of BiLSTM architectures and transformer architecture layers. The applied combination provided in-depth understanding of the text content and contributed to increase in propaganda detection by 0.037 compared to known analogues. To classify propaganda techniques, both the classification of the applied techniques and visualization of obtained results using the LIME algorithm were proposed. Set of machine learning models is used, where separate trained machine learning model based on the transformer architecture is responsible for classification of each technique, namely BERT-like models. This use allows to classify applied techniques with minimum Accuracy score of 0.82. The implemented detection of propaganda objects allows to find not only who the propaganda is directed at, but also what it is directed at in terms of classified techniques used. For interpretability and clarity, visualization of the results is also provided. Approach proposed in the work correlates with the UNDP Sustainable Development Goals and allows to automate the process of detecting and classifying propaganda and make the results of the presentation complete, interpretable and understandable. In particular, the detection and classification of propaganda techniques and objects using machine learning methods contribute to the achievement of the UN Sustainable Development Goal No. 16 by increasing the transparency of the information space and strengthening institutional trust, as well as the UN Sustainable Development Goal No. 4 by developing media literacy and critical thinking among the population, which allows to effectively counteract disinformation.