Machine Learning Method for Detecting Propaganda in Twitter Texts

This paper introduces a natural language processing and machine learning-based approach tailored for propaganda detection in textual data. Focused on the context of the 2022 Russian invasion of Ukraine, we present a novel dataset featuring recent instances of Russian Federation state propaganda and disinformation on Twitter. Our baseline machine learning classification approach for propaganda detection harnesses the Word2Vec word vectorization model, the Random Forest classifier, and incorporates text preprocessing techniques such as tokenization, lemmatization, and stop-word removal. The study emphasizes the critical role of selecting appropriate text preprocessing methods. Expanding the scope, we further extend our approach to process text in the Ukrainian language, enhancing its applicability in a multilingual context. Additionally, we conducted exploratory data analysis, oulining distribution nuances in text sizes for both English and Ukrainian datasets. The dataset and the source code are made publicly available on Kaggle, aiming to encourage further developments in combating propaganda and disinformation.