Efficient detection of online communities and social bot activity during electoral campaigns

Threats of social media manipulation during elections have become a central concern for modern democracies. This study tackles the problem of identifying the purpose and origins of social bots during electoral campaigns. We propose a methodology - uniform manifold approximation and projection combined with user-level document embeddings - that efficiently reveals the community structure of social media users. We show that this method can be used to predict the partisan affiliation of social media users with high accuracy, detect anomalous concentrations of social bots, and infer their geographical origin. We illustrate the methodology using Twitter data from the 2019 Canadian electoral campaign. Our evidence supports the thesis that social bots have become an integral component of campaign strategy for national actors. We also demonstrate how the methodology can be deployed to identify clusters of foreign bots, and we show that such accounts were used to share far-right and environment-related content during the campaign.