Detecting Propaganda in Ukraine-Russia War Tweets by Politicians: A CACDA and Computational Perspective

Social media platforms have emerged as critical tools for political communication, particularly during conflicts where information is weaponized to shape public opinion and influence political realities. The Ukraine-Russia war exemplifies this phenomenon, with Twitter (now X) functioning as a digital battlefield. On this platform, political actors employ propaganda to (de)legitimize ideologies and assert power dynamics, thereby shaping socio-political narratives. This study conceptualizes and theorizes propaganda through the lens of historical wars, propaganda theories, and Critical Discourse Analysis (CDA). It focuses on key propaganda components, including ideologies, power relations, persuasive argumentation, and intertextuality. Based on this theoretical framework, the research collected data and developed guidelines, a scheme, and an annotation platform to establish a robust, validated dataset. The dataset, collected from X (formerly Twitter), includes English-language textual tweets posted by Ukrainian and Russian politicians between January 1, 2022, and April 30, 2023. This corpus includes 1,000 tweets containing 41,959 tokens, annotated with 44 labels through a Google Form-based platform to identify propaganda components. Additionally, Twitter metrics—such as retweet count, reply count, like count, tweet count, and follower count—were analyzed quantitatively to profile the (re)contextualization of content. The data analysis employed exploratory, corpus-based, computational, and relational methods, each addressing distinct research objectives. Exploratory analysis uncovered dominant ideological patterns, notably Anti-Russian, Pro-Ukrainian, and Anti-War ideologies. The findings portray Ukraine as a primary victim and Russia as a primary oppressor, highlighting the use of emotional appeals to create a discourse that is ideological, power-oriented, and persuasive. Conversely, the CACDA (Corpus-Assisted Critical Discourse Analysis) employed Sketch Engine to examine linguistic features within the entire tweet corpus. This analysis explored the most frequently occurring grammatical categories, keywords, and multi-term keywords, subsequently narrowing them down for detailed word sketches and concordance analysis. The results reveal that grammatical categories such as lemmas, nouns, and adjectives signify actors, ideologies, and power relations, while adverbs indicate the tone of the discourse. Keywords highlight recurring themes and topics, and multi-term keywords offer deeper insights into propaganda strategies. Collocation analysis identifies structures such as “verbs with keywords as objects” and “modifiers of keywords”, consistently uncovering ideological stances, power dynamics, and persuasive elements. A comparative analysis of Anti-Russian and Non-Anti-Russian sub corpora shows that Anti-Russian tweets exhibit more negative connotations and frequently invoke emotional appeals to convey ideologies. Key nouns, adjectives, and multi-term keywords play pivotal roles in constructing these ideologies, while word sketches and concordance analyses situate these linguistic features within broader propagandistic narratives. The study also developed a multi-label classifier to predict labels related to ideologies, power relations, and persuasive techniques in unseen war-related tweets. Trained and evaluated on the annotated dataset, the classifier demonstrates variable performance across labels, with higher accuracy for frequently observed labels compared to those less common. Additionally, clustering tweets based on X’s engagement metrics revealed high, moderate, and low engagement clusters, offering insights into audience interaction with propagandistic content. To support practical analysis, the study introduced a web-based application that facilitates the sequential exploration of propaganda components while integrating engagement metrics. This tool allows users to filter tweets by labels and conduct qualitative analyses of selected examples, providing a hands-on approach to understanding the dynamics of propaganda and its dissemination. This research presents replicable frameworks for analysing digital propaganda in political discourse during conflicts. Its findings contribute to media literacy initiatives, empower policymakers to address vulnerabilities in public opinion, and assist platform moderators in identifying and mitigating propaganda in real-time.