Abstract
Traditional network models typically assume that social structures grow through the addition of individual links between pairs of nodes. However, real-world social interactions, such as scientific collaborations or professional meetings, are often group-based "events." This paper introduces the Evolution-aware Event-driven Locality and Attachedness-based Growth (EELAG) model. EELAG moves beyond edge-centric approaches by simulating the formation of cliques through events while accounting for the temporal evolution of participant behavior. Validated against large-scale nanotechnology collaboration datasets, the model accurately replicates key topological properties like high clustering, assortativity, and degree distributions.
Introduction
Most social network research relies on models like the Barabási-Albert "preferential attachment" framework, which are fundamentally edge-centric. These models assume growth occurs one link at a time. In reality, connections often form through group events—when multiple authors collaborate on a single paper, they form a "clique" where every participant is connected to every other participant.
Edge-centric models often struggle to recreate the high levels of clustering and specific assortative mixing patterns found in real social networks without adding artificial rules. Furthermore, they typically treat node behavior as static, ignoring how an individual's collaboration patterns change over their professional lifetime.
The EELAG Model
The proposed EELAG model addresses these shortcomings through several key innovations:
- Event-Driven Framework: The fundamental unit of growth is the "event" rather than the "edge." Each event involves $m$ participants who form a complete sub-graph.
- Behavioral Evolution: The model accounts for the fact that average event sizes (e.g., authors per paper) tend to increase over time. It also incorporates node "lifetimes," recognizing that many participants (such as graduate students) may only be active for a short period.
- Hybrid Selection Logic: To determine event participants, EELAG combines Attachedness (preferential attachment to popular nodes) with Locality (the tendency to connect with "friends of friends").
- Span Distance: The model differentiates between "explorers" (often junior researchers) who connect across large topological distances and "consolidators" (often senior researchers) who maintain stable local collaboration groups.
Empirical Trends in Nanotechnology
The researchers validated EELAG using two massive datasets: NanoSCI (nearly 300,000 researchers) and NanoParticle (a specialized sub-community). These datasets provide a clear record of "events" (papers) over several decades.
A critical observation from this data is the steady increase in collaboration. As shown in Figure 2, the mean number of coauthors has grown significantly as the networks have expanded, reflecting a shift toward more collaborative research practices.
Key Results and Insights
By modeling events rather than edges, EELAG naturally produces the high clustering coefficients observed in real-world data. Unlike traditional models, EELAG also successfully replicates "assortative mixing," where highly connected nodes tend to collaborate with other highly connected nodes.
The model's inclusion of node mortality is also vital. In the nanotechnology datasets, approximately 80% of researchers remain active for only a single year. By accounting for these short lifetimes and the evolving size of collaboration groups, EELAG achieves a degree distribution that closely matches the real-world power-law with an exponential cut-off.
Conclusion
The EELAG model demonstrates that an event-driven approach, combined with an understanding of how participant behavior evolves over time, provides a much more accurate representation of social network growth than traditional edge-centric models. While currently focused on scientific collaboration, this framework offers a robust foundation for studying other event-driven social structures.
Summary updated