How U.S. Presidential elections strengthen global hate networks

The study investigates how local or national offline events—specifically the 2020 U.S. presidential election and ensuing Capitol attack—drive rapid, global-scale adaptations in an interconnected “hate universe” spanning multiple social media platforms. The authors conceptualize this universe as a network-of-networks formed by hate communities (e.g., Telegram channels, Gab groups, YouTube channels) that cross-link by sharing URLs, thereby exposing members to other communities and drawing a “vulnerable mainstream” into proximity. Prior work shows contempt and in-group superiority can fuel hate speech, but a global, structural perspective on how such hate adapts around real-world triggers has been lacking. The purpose is to quantify how election-related events affect the topology (cohesion, connectivity) and the content (targets and themes) of hate networks at internet scale, and to derive implications for policy and intervention.

The paper builds on research into online hate, misinformation, and cross-platform dynamics, noting that while many studies focus on specific platforms or singular events, fewer examine global, co-evolving structures and contents across platforms. Prior work addresses the emergence of extreme media ecosystems, meme flows from fringe communities, Telegram’s role as a haven for extremist mobilization, and network visualization and control in complex systems. The authors identify a gap: understanding how a local or national event (e.g., an election) induces rapid, global reconfiguration and content shifts across a multi-platform hate network, and how smaller platforms (e.g., Telegram) play distinct structural roles often overlooked by policy.

The authors reconstruct a cross-platform “hate universe” by iteratively mapping hate communities and the links they create when cross-posting content. Initial seed communities were identified using public resources (e.g., ADL Hate Symbols Database, SPLC) and expert review. The team then monitored these communities for links (shared URLs) leading to other communities, expanding the candidate list via link-following and subject-matter-expert validation to determine hate status. The resulting directed network comprises hate communities as source nodes and their targets (which include both hate and non-hate communities, the latter labeled as “vulnerable mainstream”). Network evolution is analyzed over the period spanning the 2020 U.S. election (Nov 1, 2020 to Jan 10, 2021), with metrics including link creation rates, connectivity (triadic closure tendency), assortativity (preference for similarly connected nodes), number of communities, and largest community size. Visualization employs ForceAtlas2 and PyVis, with figures refined in Adobe Illustrator. Content analysis uses NLP models to classify posts into seven hate categories: race, gender, religion, antisemitism, gender identity/sexual orientation (GISO), immigration, and ethnicity/determinism/nationalism (EIN). Models achieved at least 91% accuracy and were validated by human subject-matter experts using reliability metrics. Platform-specific daily link counts are correlated with content category frequencies to assess structure–content coupling. Data collection spans multiple platforms (e.g., Telegram, 4chan, Gab, Twitter), producing a hate-universe that encompasses several billion individual accounts when including vulnerable mainstream communities. Processed datasets and plotting code are made available via a public GitHub; raw data are restricted due to sensitivity.

• Surge in hate links around key election events: On Nov 3 (election day), links from hate communities increased by 41.6% vs Nov 1; on Nov 7 (president-elect declared), the increase reached 68%. A larger spike occurred around Jan 6 (Capitol attack).
• Structural hardening: After Jan 6, connectivity increased by 164.8%, assortativity rose by 27%, the number of communities decreased by 19.8%, and the size of the largest community grew by 16.27%, indicating consolidation into a denser, more cohesive network-of-networks with stronger core interlinkages.
• Content hardening: Around Nov 7, anti-immigration content surged by 269.5%, ethnicity-based hatred by 87.9%, and antisemitism by 117.57% (relative to the preceding window). Around Jan 6, anti-immigration content increased by 108.6% (Jan 6–10 vs Jan 1–5). These align with Great Replacement narratives and conspiracies attributing demographic change to Jewish influence.
• Platform roles and correlations: Telegram acts as a key “glue” platform. From Nov 4–7 vs Nov 1–3, links involving Telegram increased by 29% (1922 to 2366). Telegram’s share as a target rose from 18.22% to 33.47%, and as a source from 21.73% to 37.24% within the hate-to-hate subnetwork. Correlations show rising links from 4chan, Gab, and Twitter, with Telegram strongly correlated to hate targeting immigration, race, and GISO.
• Scale and reach: Approximately 50 million accounts in hate communities were drawn closer to each other and toward a mainstream of billions, highlighting rapid global amplification potential from locally anchored events.

The findings demonstrate that offline political events can rapidly and globally intensify both the cohesion and thematic focus of online hate networks. Structural hardening (higher connectivity and assortativity, consolidation into larger components) makes the ecosystem more resilient to disruption, while content hardening concentrates attention on themes (immigration, ethnicity, antisemitism) not strictly limited to the event’s immediate topic. Telegram’s outsized binding role underscores that smaller or less-regulated platforms can be central to network consolidation and content propagation. These results address the research question by evidencing how individual-level cross-posting behaviors scale to global network reconfiguration, drawing vulnerable mainstream communities into closer proximity. Implications include the need for interventions that are cross-platform, attuned to platform-specific roles, and that blend multiple hate themes rather than focusing solely on the event’s surface narrative. Policymaking that targets only major platforms or treats all platforms uniformly is unlikely to curb the observed dynamics.

The study introduces a scalable, cross-platform mapping of a global hate network-of-networks and shows that the 2020 U.S. presidential election and Jan 6 catalyzed rapid structural and content hardening. It quantifies spikes in linking activity, increases in cohesion metrics, consolidation of communities, and surges in specific hate themes aligned with conspiratorial narratives, highlighting Telegram’s central role. For practice and policy, the authors recommend proactive, multi-platform strategies that blend messaging across hate themes and target platform-specific structures, leveraging a multi-dimensional map of the hate universe. Future research should analyze the composition and behaviors of the vulnerable mainstream in greater depth, refine causal inference between online dynamics and offline events, and explore generalization to other global events (e.g., elections, conflicts).

• Data collection across social media platforms is necessarily imperfect and incomplete, potentially biasing coverage of communities and links.
• The study focuses on topological and content changes without establishing causal impact on offline events; observed patterns may reflect responses rather than drivers.
• The vulnerable mainstream’s composition is not analyzed in detail, limiting insight into susceptibility and pathways of influence.
• Raw data cannot be shared due to sensitivity and protection standards; although processed derivatives are provided, this constrains independent validation and replication granularity.
• NLP classification, while high-accuracy and expert-validated, may still misclassify nuanced content or platform-specific vernacular.