Interest communities and flow roles in directed networks: the Twitter network of the UK riots

Directionality is a crucial ingredient in many complex networks in which information, energy or influence are transmitted. In such directed networks, analysing flows (and not only the strength of connections) is crucial to reveal important features of the network that might go undetected if the orientation of connections is ignored. We showcase here a flow-based approach for community detection in networks through the study of the network of the most influential Twitter users during the 2011 riots in England. Firstly, we use directed Markov Stability to extract descriptions of the network at different levels of coarseness in terms of interest communities, i.e., groups of nodes within which flows of information are contained and reinforced. Such interest communities reveal user groupings according to location, profession, employer, and topic. The study of flows also allows us to generate an interest distance, which affords a personalised view of the attention in the network as viewed from the vantage point of any given user. Secondly, we analyse the profiles of incoming and outgoing long-range flows with a combined approach of role-based similarity and the novel relaxed minimum spanning tree algorithm to reveal that the users in the network can be classified into five roles. These flow roles go beyond the standard leader/follower dichotomy and differ from classifications based on regular/structural equivalence. We then show that the interest communities fall into distinct informational organigrams characterised by a different mix of user roles reflecting the quality of dialogue within them. Our generic framework can be used to provide insight into how flows are generated, distributed, preserved and consumed in directed networks.

💡 Research Summary

This paper presents a flow‑centric framework for community detection and role classification in directed networks, illustrated through the analysis of Twitter activity during the 2011 UK riots. The authors began by harvesting the “most influential” 1,000 Twitter accounts listed by The Guardian, then reconstructed the directed follower network among these accounts in February 2012, yielding a largest connected component of 914 nodes and several thousand directed edges (an edge points from a follower to the account they follow, i.e., the direction of declared interest, while information flows opposite to the edge).

To uncover groups of users that retain and reinforce information flows, the authors applied Directed Markov Stability, a multiscale community‑detection method based on continuous‑time Markov diffusion. By varying the Markov time parameter t, they obtained a hierarchy of partitions: at short times (t≈0.15) 149 fine‑grained communities, at intermediate times (t≈0.5) 48, at longer times (t≈1.3) 15, and at the coarsest level (t≈7) four large communities. Each community was annotated post‑hoc with word‑clouds derived from users’ self‑descriptions, revealing interpretable themes such as geographic clusters (e.g., Manchester, Hackney), media organisations (ITV, Daily Telegraph), police and crime reporting, sports, and entertainment.

A key contribution is the explicit comparison between directed and undirected analyses. When edge directionality is ignored, prominent flow‑based groups—most notably the BBC cluster—disintegrate as the undirected Markov Stability merges them with unrelated accounts. In contrast, the directed version preserves the BBC community across all scales because its nodes have high in‑degree and PageRank, acting as strong attractors of flow. Some communities (e.g., the George Monbiot cluster) are robust to directionality, indicating balanced inbound and outbound flow.

The authors introduce “interest distance,” defined as the earliest Markov time at which a node shares a community with a chosen “vantage point.” This yields an ultrametric distance that can be visualised from any user’s perspective. Using Anonymous as a vantage point highlights proximity to WikiLeaks, Al Jazeera, and human‑rights organisations, while sports figures appear far away. Conversely, using footballer Wayne Rooney as the reference places other athletes and entertainment accounts nearby and political or activist accounts at larger distances. This personalized metric captures nuanced directed pathways that traditional centrality measures miss.

Beyond community structure, the paper develops a flow‑role classification. First, a role‑based similarity matrix is constructed from the patterns of long‑range incoming and outgoing flows for each node. The matrix is then sparsified using a relaxed minimum spanning tree (RMST), preserving the most informative similarities while discarding noise. Applying the same multiscale stability approach to this similarity graph yields five distinct flow roles: (1) information sources (high in‑degree, high PageRank, strong outward flow), (2) information sinks (low in‑degree, high out‑degree, mainly consume flow), (3) relayers (balanced inbound/outbound flow, act as bridges), (4) cross‑topic mediators (strong bidirectional flow within a thematic niche), and (5) mixed actors (exhibit multiple flow patterns). These roles extend beyond the classic leader/follower or hub/authority dichotomies by explicitly accounting for how nodes generate, transmit, and retain information.

Finally, the authors map each interest community onto a “information organigram” that displays the composition of flow roles within it. For example, the BBC community is dominated by source roles, reflecting its capacity to broadcast and retain attention, whereas the Anonymous community contains a richer mixture of cross‑topic mediators and relayers, indicating a more distributed dialogue. Such organigrams provide a novel lens for assessing the quality of discourse and the efficiency of information diffusion within sub‑networks.

Overall, the study demonstrates that (i) directed flow‑based community detection uncovers socially meaningful groupings that are invisible to undirected methods, (ii) interest distance offers a personalized view of network relevance, and (iii) flow‑role analysis yields a functional taxonomy of nodes grounded in actual information dynamics. The framework is generic and can be applied to other directed systems such as brain connectomes, trade networks, or protein interaction maps, where the direction of interaction fundamentally shapes system behaviour.