How U.S. Presidential elections strengthen global hate networks

The research explores the amplification of online hate speech on a global scale, focusing on the 2020 U.S. presidential election as a case study. The internet, with its interconnected web of hate communities across various platforms, presents a unique challenge in understanding and addressing the spread of hateful ideologies. This study defines a "hate universe"—a network-of-networks—that comprises hate communities and the mainstream communities they influence. The research question centers on how a localized event like a U.S. presidential election can trigger a global surge in online hate and strengthen the hate universe's structure and content. The importance of this study lies in understanding the dynamic interplay between offline events and the online spread of hate, informing more effective strategies to counter online hate speech. The 2020 election provides a rich context, given the emergence of new platforms and the inclusion of events such as the Capitol attack. The study aims to move beyond analyses focusing solely on individual platforms, recognizing the interconnected nature of the global hate network and its resilience to societal pressure.

Existing literature highlights the prevalence of hate speech online and its concentration within interconnected networks across various social media platforms. Studies have examined the relationship between real-world events and the evolution of online hate; however, this research focuses on the dynamics of the hate universe at a global scale to bridge this knowledge gap. Prior work has explored the structure and content of online hate communities individually, but this study leverages network analysis to understand the collective behavior of these communities and their influence on a broader mainstream. The concept of in-group superiority and its contribution to hate speech is mentioned, setting the stage for a deeper understanding of the psychological factors driving the dynamics observed in the study.

The methodology builds upon previous research in identifying and mapping hate communities across social media platforms. Hate communities are defined based on expert assessment of their content, aligning with the FBI's definition of hate crimes or promotion of extreme racial identitarianism. An initial list of communities was identified using public resources such as the ADL Hate Symbols Database and the Southern Poverty Law Center. Link-following techniques were employed to expand the list by monitoring cross-posts between communities, effectively mapping the hate universe. Natural language processing (NLP) models were trained to classify hate speech into seven categories: race, gender, religion, antisemitism, gender identity/sexual orientation (GISO), immigration, and ethnicity/determinism/nationalism (EIN). The NLP models boasted high accuracy scores (at least 91%). Data collection, as acknowledged, faced limitations, but the resulting hate universe encompassed billions of individual accounts. The analysis focuses on changes in the network topology around the 2020 U.S. presidential election and the Capitol attack. Data protection standards prevented the public release of the raw data, but the processed data is available to reproduce the study's results. Python libraries (Matplotlib and PyVis) were used for data analysis and visualization. The network visualization in Figure 3 was created using the ForceAtlas2 layout algorithm.

The study found a significant surge in hate links (links from hate communities) during the 2020 U.S. presidential election and the subsequent Capitol attack. On election day, there was a 41.6% increase compared to the previous day, and this spiked further to 68% when Biden was declared president-elect. The attack on the Capitol saw an even larger spike in link creation. Key network metrics demonstrated the hate universe's hardening, characterized by increased connectivity (164.8% jump after January 6), assortativity (27% increase), and the growth of the largest community (16.27% increase) alongside a decrease in the number of communities (19.8% decrease). This hardening indicates a strengthening of existing ideologies and increased resilience to outside intervention. The content of hate speech also shifted, with significant increases in anti-immigration (269.5% surge), ethnically-based hatred (87.9% rise), and antisemitism (117.57% escalation) around the declaration of Biden's victory and the Capitol attack. Telegram emerged as a central platform in this hardening process, showing significant increases in connections and acting as a key connector within the hate network. Strong correlations were found between changes in the hate universe's structure and its content, with Telegram exhibiting a strong correlation with hate speech targeting various identities.

The findings demonstrate the rapid adaptation and global reach of the online hate universe, responding swiftly to localized events like a U.S. presidential election. The increased connectivity and homogeneity within the network highlight its growing resilience to external pressure. The shift in hate speech content, particularly around immigration, ethnicity, and antisemitism, underscores the need to address the underlying anxieties driving these narratives. The prominent role of Telegram indicates that current policies focused on major platforms alone are insufficient. The study's findings challenge the assumption that anti-hate campaigns should only focus on the immediate themes of a real-world event. The surge in hate content related to various identities, independent of the immediate election-related themes, suggests that broader, multi-themed approaches are necessary. The organic nature of this hate universe's strengthening underscores the need for proactive and adaptable countermeasures.

This research reveals the significant impact of U.S. presidential elections on the global online hate universe, demonstrating its rapid self-adaptation and ability to amplify hate speech at scale. The study highlights the limitations of current anti-hate strategies and underscores the need for proactive measures that consider the multi-dimensional nature of the hate universe and its ability to adapt to global events. Future research should explore the specific mechanisms driving the convergence of hate communities and the interplay between online and offline dynamics in shaping the evolution of online hate.

The study acknowledges limitations inherent in social media data collection. While the hate universe analyzed encompasses billions of accounts, the data represents only a portion of online activity and may not fully capture the nuances of hate speech dynamics. Furthermore, determining causality between online hate and offline events remains a challenge. Finally, the study's focus on the 2020 U.S. election and its immediate aftermath does not guarantee the generalizability of findings to other events or contexts.