Abstract
In this paper, we propose an alternative model for collaboration networks based on nonlinear preferential attachment. Depending on a single free parameter "preferential exponent", this model interpolates between networks with a scale-free and an exponential degree distribution. The degree distribution in the present networks can be roughly classified into four patterns, all of which are observed in empirical data. And this model exhibits small-world effect, which means the corresponding networks are of very short average distance and highly large clustering coefficient. More interesting, we find a peak distribution of act-size from empirical data which has not been emphasized before. Our model can produce the peak act-size distribution naturally that agrees with the empirical data well.
Introduction
In network science, the Barabási-Albert (BA) model is the standard for explaining the "rich get richer" phenomenon through linear preferential attachment, leading to scale-free networks. However, real-world collaboration networks—where nodes represent individuals and edges represent shared "acts" like research papers or movie casts—often deviate from this mold. Empirical data shows a spectrum of degree distributions, from exponential to power-law. Furthermore, most models treat "act-size" (the number of participants in a single collaboration) as an external constant. This research proposes a unified growth model governed by nonlinear preferential attachment that naturally accounts for both diverse degree distributions and emergent act-size patterns.
Methodology
The proposed model evolves through a growth process where a new node is added at each time step. This node initiates collaborations with existing nodes based on a nonlinear selection rule. The probability $\pi(k)$ that an existing node with degree $k$ is chosen is proportional to $k^\alpha$, where $\alpha$ is the preferential exponent.
Unlike models with a fixed number of edges per step, this model introduces a parameter $\lambda$ to control the average number of collaborators. The actual number of nodes chosen in any given step is probabilistic, leading to a dynamic and emergent "act-size." To analyze the resulting topologies, the authors utilize the Stretched Exponential Distribution (SED), a versatile mathematical form that can describe a wide range of connectivity patterns by adjusting a single shape parameter $c$.
Results and Discussion
Connectivity and Degree Distribution
By tuning the preferential exponent $\alpha$, the model interpolates between four distinct connectivity patterns: Exponential ($\alpha=0$), a transitional Arsy-Varsy form, Semi-Power Law, and Power Law ($\alpha \approx 1.5$). This suggests that different real-world networks are not fundamentally different species but rather exist at different points on the $\alpha$ spectrum.
As shown in Figure 4, the degree distribution fits the SED remarkably well. As $\alpha$ increases, the SED parameter $c$ decreases, indicating a shift from near-exponential behavior toward heavy-tailed power laws. This flexibility allows the model to match empirical datasets ranging from scientific co-authorship to tourism routes.
Small-World Metrics
The model successfully replicates the "small-world" phenomenon, producing networks with high clustering coefficients and very short average path lengths. For a network of 5,000 nodes, the clustering coefficient $C$ remains high (above 0.4), a hallmark of social collaboration that standard BA models often fail to replicate.
Figure 1 demonstrates that the average path length $L(N)$ scales logarithmically with the number of nodes $N$ ($L \sim \ln N$). Interestingly, while the clustering coefficient is sensitive to the preferential exponent $\alpha$, it is largely unaffected by the average act-size $\lambda$, suggesting that the intensity of preferential attachment is the primary driver of local connectivity.
Emergent Act-Size Distribution
A key contribution of this model is its ability to naturally produce a single-peaked, exponentially decaying distribution for act-sizes. Empirical data from Physical Review Letters shows that most papers have a typical number of authors (e.g., 3-5), with very few having exceptionally large counts. The model reproduces this "peaked" behavior without it being explicitly hard-coded, matching empirical observations across various social and logistical datasets.
Conclusion and Future Work
This nonlinear preferential attachment model provides a robust, unified framework for understanding collaboration networks. While simple, it captures the essential growth dynamics that lead to diverse degree distributions and realistic act-size patterns. Future research could extend this work by incorporating node aging, explicit community structures, and weighted edges to better model the frequency and longevity of real-world collaborations.
Summary updated