Investigation of machine learning approaches to classify war-related content during Russian full-scale invasion of Ukraine

Social media platforms are digital arenas for information dissemination. Consequently, they have become a battlefront for disinformation, propaganda, fake news, and strategic narratives during emerging geopolitical events such as war. Machine learning has increasingly been developed to systematically analyze and classify digital content, shedding light on online propaganda’s underlying patterns and strategies. Here we applied multiple machine learning algorithms to classify pro-Russian communications on Twitter (tweets) following the Russian full-scale invasion of Ukraine. Machine learning models included Logistic Regression, Support Vector Machine, Bi-directional Long Short Term Memory, Naive Bayes, K-Nearest Neighbours, and Extreme Gradient Boosting. Model performances were evaluated based on accuracy, precision, recall, and F1-score metrics. The Support Vector Machine and Extreme Gradient Boosting models consistently outperformed others, achieving higher accuracy and F1 scores. In general, model performance improved with increasing dataset size. The results highlighted the complexities of online informational manipulation, emphasizing the need for a deeper sentiment analysis. This study offers a pioneering contribution to understanding information manipulation in the context of the Russia-Ukraine war and provides valuable insights into the intricacies of digital information warfare.