Abstract
Standard social network models often fail to capture the unique dynamics of mentor-apprentice relationships in online environments. While typical networks exhibit shrinking diameters and coalesce into a single giant component, mentoring networks in Massively Multiplayer Online (MMO) games show increasing diameters and maintain numerous disconnected clusters. This paper introduces the Generative Temporal Preferential Attachment (GTPA) model, which incorporates active subgraphs and a preference for younger mentors to accurately simulate these unconventional growth patterns.
Introduction
In network science, "universal laws" typically dictate that as a network grows, its diameter shrinks and it tends toward a single "giant component." However, empirical analysis of mentor-apprentice networks in MMOs reveals a departure from these norms. These networks, characterized by veteran players assisting novices, maintain many disconnected clusters and exhibit an increasing diameter over time. Traditional models like standard Preferential Attachment or Forest Fire models cannot account for this behavior. The challenge lies in developing a generative model that captures the specialized, exchange-based social structures of these communities.
Methodology: Triadic Census and GTPA
To understand the nature of mentoring, the researchers first conducted a triadic census, analyzing the 16 possible configurations of three-node connections. This analysis aimed to distinguish between "restricted" (one-on-one) and "generalized" (communal) mentoring.
The findings indicated that mentoring is primarily a series of specialized, dyadic (restricted) exchanges. Based on this, the GTPA model was proposed, modifying standard Preferential Attachment with three key mechanisms:
- Component Creation: New nodes join in small, initially isolated groups.
- Active Graph Assumption: Only a subset of the population is "active" and available for mentoring at any given time.
- Temporal Preferential Attachment: New nodes prefer connecting to younger nodes (recent joiners), reflecting the "window of opportunity" for mentorship.
Dataset: EverQuest II
The study utilized a massive temporal dataset from the fantasy MMO EverQuest II (EQ2), covering an eight-month period in 2006. The researchers tracked explicit "mentoring" links—a specific game mechanic where high-level players temporarily lower their level to quest with lower-level players. Data was analyzed at various temporal granularities, from hourly to monthly.
Key Results and Insights
The GTPA model successfully replicates the structural anomalies observed in the EQ2 data. Unlike standard models, GTPA captures the increasing diameter and the growing number of components over time.
Breaking the "Shrinking Diameter" Law
Mentoring networks expand and become sparser as new, isolated pockets of mentorship emerge. This is driven by the "youth preference": apprentices seek mentors who are currently active and available, rather than just the most established players. Consequently, the model weights node age inversely.
Structural and Triadic Accuracy
The GTPA model's triadic census closely matched the real-world distribution, confirming it captures the local social "flavor" of mentoring. Furthermore, while the Largest Connected Component (LCC) grows, it does not dominate the entire network; a significant portion of the population consistently remains in smaller, secondary clusters.
Conclusion and Future Work
The GTPA model provides a robust framework for understanding specialized social exchanges in online games. However, it currently produces monotonic growth, missing the "bursty" behavior seen in real-world data (e.g., weekend spikes). Future work could incorporate informal help-giving (chat/forums) and explore how game incentive designs might shift these networks from dyadic to more communal structures.
Summary updated