Introduction
The migration of extremist groups from dedicated websites to mainstream social media platforms like Twitter has fundamentally altered their reach and recruitment strategies. Far from being a monolithic entity, the "extreme right" encompasses a diverse array of factions—including white supremacists, ultra-nationalists, and anti-EU activists—with overlapping yet distinct ideologies. This research maps these digital ecosystems to investigate whether these groups operate in national silos or engage in cross-border interactions, and whether stable sub-communities can be identified through their discourse.
Methodology and Data
The study utilized a multi-step computational approach on Twitter data collected between March and August 2012. An initial set of 275 accounts across eight countries (France, Germany, Greece, Italy, Spain, Sweden, UK, and USA) was expanded through reciprocal relationships to create larger case studies: an English language dataset (1,513 accounts) and a German language dataset (448 accounts).
Researchers constructed two types of undirected networks:
- Reciprocal Follower Networks: Representing passive awareness between accounts.
- Reciprocal Interaction Networks: Based on mentions and retweets, representing active dialogue.
To identify clusters, the OSLOM algorithm was applied to detect "consensus communities"—groups with significantly higher internal interaction than external. To interpret the ideological "flavor" of these communities, Non-negative Matrix Factorization (NMF) was applied to hashtags. NMF proved superior to Latent Dirichlet Allocation (LDA) for this niche data, providing more specific and interpretable labels by mapping hashtag usage to discovered topics via cosine similarity.
International Connectivity and the Role of English
A primary finding of the research is the role of the English language as an extremist lingua franca. While most interactions are geographically and linguistically localized, English serves as a critical bridge. Swedish and Italian accounts tweeting in English often act as central "hubs," connecting European factions to those in the United States.
The network analysis reveals a high level of international awareness through follower links, as seen in the interconnectedness of the follower network. However, there is a distinct difference when examining actual interaction.
Awareness vs. Active Interaction
While extremists monitor their international counterparts, their active dialogue remains largely parochial. The interaction network is significantly more fragmented than the follower network, suggesting that while there is a global "passive" community, active collaboration is often restricted by national borders and language.
Ideological Diversity and Sub-Communities
The combination of community detection and NMF successfully isolated distinct sub-groups. In the English-speaking network, clusters were identified for the English Defence League (EDL), White Power (WPWW), and traditional conservatives (TCOT). In Germany, the analysis revealed a geographic split within the NPD (National Democratic Party) between Western and Eastern branches.
A significant thematic bridge identified was "counter-Jihad" sentiment. This ideology often served as common ground, linking more moderate nationalist groups with extreme white supremacist factions across different nations.
Conclusion and Future Directions
The study demonstrates that automated community detection and topic modeling can capture the vast majority of active participants in extremist networks (86% in the English network and 74% in the German network). However, the research is limited by its reliance on a static snapshot from 2012 and potential sampling bias from the initial seed accounts.
Future work should explore the temporal evolution of these networks—how they form, merge, or radicalize over time—and utilize more granular probabilistic modeling to uncover deeper sub-topics within these ideological clusters.
Summary updated