Characterizing interactions in online social networks during exceptional events

Nowadays, millions of people interact on a daily basis on online social media like Facebook and Twitter, where they share and discuss information about a wide variety of topics. In this paper, we focus on a specific online social network, Twitter, and we analyze multiple datasets each one consisting of individuals’ online activity before, during and after an exceptional event in terms of volume of the communications registered. We consider important events that occurred in different arenas that range from policy to culture or science. For each dataset, the users’ online activities are modeled by a multilayer network in which each layer conveys a different kind of interaction, specifically: retweeting, mentioning and replying. This representation allows us to unveil that these distinct types of interaction produce networks with different statistical properties, in particular concerning the degree distribution and the clustering structure. These results suggests that models of online activity cannot discard the information carried by this multilayer representation of the system, and should account for the different processes generated by the different kinds of interactions. Secondly, our analysis unveils the presence of statistical regularities among the different events, suggesting that the non-trivial topological patterns that we observe may represent universal features of the social dynamics on online social networks during exceptional events.

💡 Research Summary

This paper investigates how users interact on Twitter during six distinct “exceptional events” – the 2013 Cannes Film Festival, the 2012 Higgs boson discovery, the 2013 Martin Luther King “I Have a Dream” anniversary, the 2013 Moscow World Athletics Championships, the 2014 New York People’s Climate March, and the 2013 Obama visit to Israel. For each event the authors collected all tweets containing predefined keywords or hashtags using Twitter’s streaming API (supplemented by a small fraction of search‑API data). From the raw stream they extracted three elementary interaction types: retweets (RT), mentions (MT) and replies (RP). Each interaction type was represented as a directed edge in a separate layer of a multilayer network, while all layers shared the same set of nodes (the users who participated in the event). Consequently, each event is modeled as a three‑layer directed multiplex network with layer‑specific edge sets.

The study addresses three core questions: (1) Do the three interaction types generate distinct network topologies within the same event? (2) Are there universal structural patterns across different events despite their heterogeneous temporal profiles? (3) What are the implications for modeling online social dynamics?

To answer (1) the authors compute an edge‑overlap metric oαβ = |Eα ∩ Eβ| / min(|Eα|,|Eβ|) for every pair of layers (α,β). Across all events the overlap values are low (≈0.05–0.08), indicating that the pairs of users who retweet each other are largely different from those who mention or reply to each other. Thus, each interaction type captures a largely independent channel of communication.

For (1) they also examine degree‑degree correlations across layers. Using the in‑degree kᵢ,α (the number of incoming interactions a user receives on layer α) they calculate Spearman rank correlations between each pair of layers. The correlations are modest (≈0.05–0.35), with the highest value observed between replies and retweets. This shows that a user who is highly retweeted is not necessarily highly mentioned or replied to, confirming that centrality is layer‑specific.

Regarding (2), the authors analyse the degree distributions of each layer. All layers exhibit heavy‑tailed distributions (approximately power‑law or log‑normal), but the retweet layer consistently shows a steeper tail, reflecting a pronounced “core‑periphery” structure where a few users attract a disproportionate amount of attention. Mention and reply layers have flatter tails, suggesting more egalitarian interaction patterns. Clustering coefficients also differ across layers, with retweets typically displaying higher clustering.

Despite the wide variation in tweet volume over time (some events last a few hours, others span weeks), the structural signatures—low edge overlap, modest cross‑layer degree correlation, and heavy‑tailed degree distributions—are remarkably consistent across all six events. This points to universal mechanisms governing how people use different Twitter functionalities during high‑visibility, time‑limited episodes.

The paper concludes that single‑layer representations of Twitter activity are insufficient for capturing the richness of user behavior. Multilayer models are essential for realistic simulations of information diffusion, influencer identification, and opinion dynamics. Moreover, the observed regularities suggest that the same multilayer structural patterns may arise in other domains where sudden spikes of online activity occur (e.g., political crises, natural disasters, viral memes). Future work could explore temporal evolution of layers, inter‑layer coupling mechanisms, and the integration of additional interaction types (e.g., likes, quote‑tweets) to further refine our understanding of online social dynamics.