Uncovering the Wider Structure of Extreme Right Communities Spanning Popular Online Networks

Derek O’Callaghan, Derek Greene, Maura Conway, Joe Carthy, Padraig

Given Twitter’s facility for the dissemination of content, we investigated its potential to act as one possible gateway to the wider online extreme right milieu, which spans multiple platforms such as Twitter, Facebook and YouTube in addition to other websites and blogs, etc. Our analysis focused on two case studies, pertaining to English and German language online communities. We inferred network representations to capture the relations between four diff erent extreme right online entities: (a) Twitter accounts, (b) YouTube channels, (c) Facebook profi les and (d) all other websites. As online communities are inherently dynamic, we also tracked the topology of the network temporally over a series of time steps, representing snapshots of the nodes and edges at successive intervals. We identified ed communities of nodes in each step, and matched these to construct dynamic community timelines across multiple steps. This method was used to detect persistent dynamic communities, which were present across all time steps. We selected four such communities for detailed analysis.

Of the selected English language communities, the first was most closely associated with the English Defence League (EDL), while also including nodes linked with the British Freedom party and Casuals United. The second English language community was primarily associated with the BNP, with nodes from other groups such as Combined ex-Forces (CxF), Infi dels and The British Resistance also present. The first selected German language community was most closely associated with a variety of non-electoral groups, including auBerpar-lamentarischer Widerstand (non-parliamentary resistance) entities, Freies Netz (neo-Nazi collectives), along with various ”information/news portals”. Two organizations banned by the German authorities in 2012, namely Spreelichter and Besseres Hannover, were also present. The second German language community was electoral in nature, associated with the Nationaldemokratische Partei Deutschlands (NPD), and included nodes representing regional NPD offices and individual politicians.

As the online presence of these groups extends beyond individual platforms such as Twitter or Facebook, our proposed network representation permits the structure of these wider communities to be revealed, which would otherwise not be evident if analysis was restricted to a single platform. We illustrate this by visualizing a single step community belonging to the persistent BNP dynamic community. Figure 1 contains diagrams for (a) the community with all nodes (left), and (b) non-Twitter nodes filtered (right). Filtering results in the loss of important central nodes such as the official BNP website and YouTube channel.The contrast between both figures illustrates the important linking role played by non-Twitter nodes, as the observable network is disconnected when they are not considered. In addition, the use of heterogeneous nodes enables us to immediately identify this community as being associated with the BNP.


Fig. 1. BNP community. Filtering non-Twitter nodes (right) results in central node loss and disconnection (grey=Twitter, blue=Facebook, orange=YouTube, red=other websites).

We also characterized the selected communities with a detailed analysis of the member nodes; in particular, those associated with external (non-Twitter) websites. Here, we restricted discussion to known extreme right groups and their affiliates. For each community, we provided two alternative top ten rankings of the website nodes; the first ranks the nodes in terms of their frequency of step membership, while the second ranks them in terms of their total degree across all steps, normalized with respect to the total number of steps. In both cases, the membership ranking distribution tended to be positively skewed, with a set of core community members assigned in the majority of steps, accompanied bya larger set of peripheral members who were assigned intermittently.

Figure 2 presents the two rankings of website nodes for the EDL community. As might be expected, the frequency ranking includes the official EDL and British Freedom websites. Other websites and blogs affiliated with the EDL are also present, including a Casuals United blog. Of similar interest is a blog that appears to be a mocking reference to the established anti-extremist organization, HOPE not hate. The degree ranking produces broadly similar results,with notable exceptions being the appearance of mainstream media websites. An analysis of the original URLs finds them to be associated with populist newspaper articles on topics such as the existence of UK sex grooming gangs, Muslim integration, and immigration in general. Other articles that are interpreted as perceived media persecution of groups such as the EDL are also promoted. Looking at the non-website nodes, official EDL Twitter and Facebook accounts can be observed, along with those purporting to be affiliates. Similar YouTube channels are also present, including a British Freedom channel that was suspended in late 2012.


Fig. 2. EDL community website node rankings. Membership frequency (left) and normalized degree (right).

We also analyzed related Twitter activity. Based on knowledge of external online events associated with these communities, we generated three month plots of the daily total tweet counts for (a) the accounts assigned in the corresponding two-week step community and (b) the remaining accounts in the data set. Here, we found that peaks in tweeting activity by the community accounts may be associated with external events. For example, the German non-electoral community plot in Figure 3 highlights increased activity around the times of street demonstrations, with other events such as the Besseres Hannover ban (initially banned by the German authorities; their account was subsequently blocked by Twitter within Germany) having a similar impact. Further details of the community structure, characterization and Twitter activity for all selected persistent dynamic communities, along with details of the associated data sets can be found in the paper.


Fig. 3. German non-electoral community tweet activity from August to November 2012.

In general, the membership composition of the selected communities appears to broadly correspond with contemporary knowledge of these groups; for example, the distinction between the EDL and the BNP. To a certain extent, we also see divisions according to the four-fold typology suggested by Goodwin et al., where these communities could be characterized by three of the four types; organized political parties (BNP, NPD), grassroots social movements (EDL) and smaller groups (German non-electoral). However, it could also be suggested that some overlap between these types can occur, for example, the presence of the Infi dels and British Resistance in the BNP community, or NPD- affiliated nodes in the German non-electoral community. In addition, the fi nding of Bartlett et al. that online supporters of groups such as the EDL are more likely to demonstrate than the national average may partly explain the increased levels of community Twitter activity at the time of external protests. It is worth pointing out here that using Twitter to infer structure within the wider online network of the extreme right may also introduce incompleteness, where the network representations are dependent on the initial identi fication of relevant profi les.

In our dynamic community analysis of both data sets, we found that the individual step networks tended to exhibit a considerable level of volatility. Although this may be due in part to data incompleteness, it is possible that such volatility may simply be a feature of the online extreme right presence. It is likely that this corresponds to factors such as the continual emergence of new groups, while events such as the jailing of the EDL’s Tommy Robinson or a potential ban of the NPD are also likely to impact volatility. We would like to address this issue in future work. In addition, it would be worthwhile to investigate the potential for other social media platforms such as Facebook to act as gateways to online extreme right activity.

For full details of the analysis, see Derek O’Callaghan, Derek Greene, Maura Conway, Joe Carthy, Padraig Cunningham, ‘Uncovering the Wider Structure of Extreme Right Communities Spanning Popular Online Networks’. WebSci’13, May 1 – May 5, 2013, Paris, France.

This entry was posted in Analysis and tagged , . : . Bookmark the permalink. Both comments and trackbacks are currently closed.
  • Navigate to regional portals

    North America Middle East South Asia Europe
  • Latest analysis from our experts

    Paul Kamolnick
    Paul Kamolnick · United States
  • Get our Extremism Tracker Email

    * indicates required