The role of hidden influentials in the diffusion of online information cascades

In a diversified context with multiple social networking sites, heterogeneous activity patterns and different user-user relations, the concept of “information cascade” is all but univocal. Despite the fact that such information cascades can be defined in different ways, it is important to check whether some of the observed patterns are common to diverse contagion processes that take place on modern social media. Here, we explore one type of information cascades, namely, those that are time-constrained, related to two kinds of socially-rooted topics on Twitter. Specifically, we show that in both cases cascades sizes distribute following a fat tailed distribution and that whether or not a cascade reaches system-wide proportions is mainly given by the presence of so-called hidden influentials. These latter nodes are not the hubs, which on the contrary, often act as firewalls for information spreading. Our results are important for a better understanding of the dynamics of complex contagion and, from a practical side, for the identification of efficient spreaders in viral phenomena.

💡 Research Summary

The paper tackles the problem of defining and characterizing information cascades on modern social media, where the notion of a cascade varies widely across studies. The authors propose a “time‑constrained” cascade definition that does not rely on exact content replication but on sequential user activity within a fixed time window (Δτ = 24 h). Starting from a seed tweet at time t₀, all followers receive the message instantly; any follower who reacts (reply, retweet, or posts a new tweet on the same topic) within the next Δτ becomes a new spreader, and the process repeats recursively. This yields a diffusion tree that includes both active spreaders and passive listeners (leaves).

Two empirical datasets from Twitter are used: (i) the Spanish “grassroots” movement (≈1.19 M tweets, 115 k users) and (ii) the 2011 Spanish elections (≈0.61 M tweets, 84 k users). For each dataset the follower network is extracted and treated as a static directed graph (edge i→j means j follows i, so information flows i→j).

The authors first confirm that cascade sizes follow a heavy‑tailed distribution: only a tiny fraction (<5 %) of cascades reach system‑wide proportions, consistent with prior work. They then examine how the structural position of the seed influences cascade growth. While out‑degree (k) of the seed shows a modest positive correlation with cascade size, the k‑core index of the seed exhibits a much stronger relationship, indicating that seeds embedded deep in the network core are far more likely to generate large cascades.

A central contribution is the identification of “hidden influentials.” These are nodes that are not hubs (i.e., they do not have the highest follower counts) but possess a high within‑module z‑score (significantly more internal links than the module average) and a high participation coefficient (many links to other modules). In other words, they act as bridges between communities while being well‑connected locally. When such nodes serve as seeds, cascades frequently become large; conversely, traditional hubs often act as firewalls because their many followers rarely react within the Δτ window, limiting further propagation.

Community structure is uncovered using the Walktrap algorithm, which optimizes modularity Q via random walks. For each node the authors compute the z‑score of internal degree and the participation coefficient Pᵢ, following Guimerà & Amaral’s framework. Nodes with high z‑score are “local hubs,” while those with high Pᵢ are “connectors.” The empirical analysis shows that successful cascades are typically initiated by connector nodes that also belong to high‑k‑core shells.

Temporal analysis reveals that cascades can persist for many days, and larger cascades tend to have longer lifetimes (temporal penetration). Topological penetration, measured as the shortest path from the seed to the farthest node in the cascade, also correlates with cascade size, confirming that deep diffusion across the network topology is essential for large‑scale spread.

Overall, the study demonstrates that in complex contagion on Twitter, structural core position (k‑core) and multi‑module bridging (high participation coefficient) are more predictive of viral spread than raw degree. This insight has practical implications: marketers, political campaigners, and public‑health communicators should target hidden influentials—users who are well‑embedded in their community yet maintain diverse cross‑community ties—rather than focusing solely on obvious high‑follower accounts. The work also provides a robust methodological framework for analyzing time‑constrained cascades, which can be applied to other platforms and topics.