AI-Driven Topic Modelling of Social Media in European Election Campaigns: Political Discourse and Voter Sentiment in Greece and Cyprus

The rapid evolution of digital technologies and the rise of artificial intelligence (AI) have fundamentally reshaped political discourse and voter engagement. Social media platforms play an increasingly significant role in shaping public opinion, influencing political processes and driving electoral outcomes. This study investigates the relationship between political discourse on social media and voter sentiment, focusing on social media data from Greece and Cyprus gathered in the framework of European Election Campaigning 2024. By analyzing datasets from digital political campaign - including social media posts and respective user reactions (such as likes, happy and sad emojis) - the research aims to uncover patterns in how political topics on social media elicit emotional responses that influence voter behaviour. The study aims to group the posts in thematic groups (political topis) by leveraging advanced digital tools such as such as Latent Dirichlet Allocation (LDA) and BERTopic, highlighting the political topics that generate the strongest emotional reactions from voters. By grading topics based on the volume of reactions they receive, the study aims to assess their influence on voter engagement. LDA, as a probabilistic generative model that identifies topics by analyzing word co-occurrences and by assigning probabilities for each word belonging to a topic, used for its simplicity and interpretability, is well-suited for discovering overarching themes that might span across multiple datasets. In contrast, BERTopic leverages pre-trained transformer-based models and large language models (LLMs) to generate contextual embeddings of text, allowing it to capture semantic nuances and dynamically adapt to contextual variations within the data. This dual approach ensures a robust analysis that balances traditional statistical techniques with cutting-edge AI-driven methods, enabling a comprehensive exploration of a variety of the important political topics discussed throughout the election period, as well as how those topics affect voters’ emotion and as a result voters’ engagement and behaviour throughout the campaign. By integrating these techniques, the study can identify both static patterns and dynamic, context-sensitive themes, thereby deepening the understanding of the thematic structure within social media posts of election campaigns. This study highlights the potential use of advanced digital tools, such as topic modeling techniques, in analyzing the complex relationship between political discourse on social media and voter engagement. By leveraging both traditional and AI-driven methodologies, the research provides valuable insights into how political topics evolve and engage voters emotionally across different contexts. The integration of temporal analyses and predictive modeling in future research will further enrich the understanding of the dynamics between online political discourse and voter behavior. This work contributes to the growing body of literature on digital democracy and the role of AI in shaping political engagement and public opinion in the digital age.