In traditional social network analysis, two processes are often studied in isolation: information diffusion (how content spreads) and network evolution (how new links form). Standard models of network growth typically rely on structural mechanisms like triadic closure—the idea that you follow a "friend of a friend." However, research by Weng et al. and Antoniades suggests that in modern social media, the content itself is the primary catalyst for structural change.
This perspective views the social graph not as a static backdrop, but as a dynamic entity shaped by a continuous feedback loop: the network structure dictates how information flows, and that flow, in turn, reshapes the network.
The Traffic-Based Shortcut
The core intuition behind this co-evolutionary view is the "traffic-based shortcut." As Weng et al. demonstrate, users frequently establish connections to content sources after encountering their posts through intermediaries (e.g., retweets). These links allow the network to bypass traditional social proximity constraints.
Antoniades formalizes this as the "Tweet-Retweet-Follow" (TRF) mechanism. When a user sees a retweet, they are exposed to a source they do not yet follow. If the content is valuable, they may follow that source directly. This process transforms loose, directed cycles of information into dense, highly connected cliques, significantly increasing the network's efficiency at spreading future information.
Identity-Revealing Hawkes Processes
To model this mathematically, the authors of the COEVOLVE framework (Farajtabar et al., Farajtabar et al.) utilize multivariate Hawkes processes. A standard Hawkes process is "self-exciting"—an event (like a tweet) increases the probability of future events (like retweets).
However, to capture network evolution, the model must be identity-revealing. Instead of just tracking the volume of activity, Farajtabar et al. argue that the model must explicitly track the original source of the information. By attributing a cascade to its origin, the model can represent how a specific user’s perceived informativeness drives their ability to attract new followers. Each time a user is exposed to a source's content, the "intensity" or rate at which they might follow that source increases.
Link Creation as a Survival Process
In the COEVOLVE framework, link creation is modeled as a survival process coupled with the diffusion process. The intuition is as follows:
- Exposure: A user is exposed to a source's content multiple times via their existing neighbors.
- Excitation: Each exposure increases the "hazard rate" of a new link forming between the user and the source.
- Termination: Once the link is established, the survival process for that specific pair "dies" (since a link can only be created once), and the new edge becomes a permanent conduit for future diffusion.
This approach, as detailed by Farajtabar et al., ensures that network evolution is directly driven by the underlying information dynamics rather than being treated as a random or purely structural event.
Behavioral vs. Structural Influence
Shifting the focus from structure to flow changes how we identify influential users. While traditional metrics like PageRank or in-degree focus on a user's position in the graph, Chen et al. propose metrics like User Bridging Magnitude (UBM), which tracks the actual paths of retweets in real-time. This captures "bridges"—users who may have fewer followers but are highly efficient at triggering deep and rapid information spread across different communities.
However, this co-evolutionary growth has limits. Ver Steeg et al. point out that high network clustering can eventually "trap" information in local groups, and "friend saturation" means that repeated exposure to the same information eventually yields diminishing returns in terms of new adoptions or links.
By treating diffusion and evolution as intertwined, continuous-time processes, researchers like Leskovec and Myers and Farajtabar et al. provide a more realistic lens for understanding how social platforms grow, how echo chambers form, and how the "bursty" nature of our conversations fundamentally rewires our social world.
Key Foundational Work
The Role of Information Diffusion in the Evolution of Social Networks (2013) — contributed 2 concept elements to this direction.
The bursty dynamics of the Twitter information network (2014) — contributed 2 concept elements to this direction.