Explosive Growth in Large-Scale Collaboration Networks

Peter R. Williams \affilnum 1,2 and Zhan Chen \affilnum 3 · 2025

Read Explosive Growth in Large-Scale Collaboration Networks by Peter R. Williams \affilnum 1,2 and Zhan Chen \affilnum 3 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-11109.txt.

Introduction

Peter Williams and Zhan Chen chart the explosive growth of two giant collaboration networks, in this 2025 paper. One maps academic co-authorship from 1800 to 2020, with hundreds of millions of nodes. The other maps film credits across a similar span.

Both grew faster than linear, following power laws that steepened after 1950 for academia. The waiting time between a person's collaborations also followed a scale-free pattern that softened over the decades. Academic teams swelled from about one to nearly six authors per paper, while film crews grew more modestly. The authors argue network models must add acceleration and coupled timescales.

Catalog record

CREATORPeter R. Williams \affilnum 1,2 and Zhan Chen \affilnum 3
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE56 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-11109

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Explosive Growth in Large-Scale Collaboration Networks Peter R. Williams \affilnum 1,2 and Zhan Chen \affilnum 3 Abstract We analyse the evolution of two large collaboration networks: the Microsoft Academic Graph (1800-2020) and Internet Movie Database (1900-2020), comprising 2.72 × 10 8 2.72\times 10^{8} and 1.88 × 10 6 1.88\times 10^{6} nodes respectively. The networks show super-linear growth, with node counts following power laws N ⁡ ( t ) ∝ t α N(t)\propto t^{\alpha} where α = 2.3 \alpha=2.3 increasing to 3.1 3.1 after 1950 (MAG) and α = 1.8 \alpha=1.8 (IMDb). Node and edge processes maintain stable but noisy timescale ratios ( τ N / τ E ≈ 2.8 ± 0.3 \tau_{N}/\tau_{E}\approx 2.8\pm 0.3 MAG, 2.3 ± 0.2 2.3\pm 0.2 IMDb). The probability of waiting a time t t between successive collaborations was found to be scale-free, P ⁡ ( t ) ∝ t − γ P(t)\propto t^{-\gamma} , with indices evolving from γ ≈ 2.3 \gamma\approx 2.3 to 1.6 1.6 (MAG) and 2.6 2.6 to 2.1 2.1 (IMDb). Academic collaboration sizes increased from 1.2 1.2 to 5.8 5.8 authors per paper, while entertainment collaborations remained more stable ( 3.2 3.2 to 4.5 4.5 actors). These observations indicate that current network models might be enhanced by considering accelerating growth, coupled timescales, and environmental influence, while explaining stable local properties. keywords Network Evolution, Collaboration Networks, Super-linear Growth, Timescale Coupling, Social Networks, Historical Analysis † † runninghead: Williams and Chen † † affiliation: \affilnum 1Rinna KK, Tokyo, Japan \affilnum 2Independent Researcher \affilnum 3Microsoft Japan, Tokyo, Japan 1 Introduction Collaboration networks are a fundamental part of human society. These networks connect people through various types of relationships, from friendship and professional collaboration to online interactions. Network science provides mathematical tools to analyse these complex systems of connections ( Newman 2018 ) . The structure and evolution of these networks have been studied extensively. Most understanding comes from examining networks at specific points in time. This snapshot approach has revealed several universal properties across different types of networks. Among the commonly observed properties are the small-world phenomenon ( Watts and Strogatz 1998 ) , the presence of power-law degree distributions ( Barabási 2016 ) , and high clustering coefficients ( Holme and Saramäki 2019 ) . The small-world phenomenon describes two key network features that can be precisely quantified. The first is a short average path length L , defined as the mean number of steps along the shortest paths between all pairs of nodes. In most real networks, L scales logarithmically with network size ( Watts and Strogatz 1998 ) . The second feature is a high clustering coefficient C , which measures the probability that two neighbours of a node are themselves connected. Typically, C is orders of magnitude larger than would be expected in random networks of the same size ( Newman and Park 2003 ) . The degree distribution P(k) of a network shows the probability that a randomly chosen node has k connections. In many real networks, this distribution follows a power law: P(k)\propto k^{-\gamma} , where \gamma typically falls between 2 and 3 ( Broido and Clauset 2019 ) . Recent work has shown that identifying true power laws requires careful statistical analysis ( Voitalov et al. 2019 ) : many networks previously thought to follow power laws may be better described by other heavy-tailed distributions ( Seshadri et al. 2021 ) . Recent theoretical work has highlighted the possibility of non-linear and explosive growth in networks. The concept of explosive percolation, introduced by Achlioptas et al. 2009 , showed that networks can undergo sudden transitions in connectivity. This was extended by studies of preferential attachment with fitness ( Bianconi and Barabási 2001 ; Dorogovtsev et al. 2000 ) , which showed how node heterogeneity can lead to accelerated growth patterns. Krioukov et al. 2010 showed how underlying geometric structures can influence network growth dynamics, potentially leading to different scaling properties. Empirical studies have found evidence for non-linear growth in various contexts. Leskovec et al. 2007 observed densification and faster-than-linear edge growth in several real networks. Kumar et al. 2010 documented similar phenomena in online social networks. However, most empirical studies cover relatively short time periods, making it difficult to distinguish between transient acceleration and fundamental growth patterns. Current models attempt to explain how these network properties emerge. The most widely accepted is preferential attachment ( Barabási and Albert 1999 ) , where new nodes prefer to connect to highly connected existing nodes. This model has been refined extensively ( Holme 2019 ) . Recent variations include fitness-based attachment ( Kong et al. 2019 ) , temporal preferences ( Ubaldi et al. 2020 ) , and memory effects ( Zhang et al. 2021 ) . Recent work has highlighted the importance of carefully examining assumptions about network growth patterns. Leskovec et al. 2007 showed that many real networks show systematic deviation from constant growth rates, while Mitzenmacher 2004 provided theoretical foundations for understanding accelerating growth processes. The temporal aspects of network evolution have received increased attention, with Karsai et al. 2011 and Kossinets and Watts 2006 highlighting the complex interplay between different timescales in social networks. These models make several important but rarely tested assumptions: 1. Steady Growth : Networks are assumed to grow at a constant or slowly varying rate. Recent work suggests this may not hold for online social networks ( Ribeiro 2014 ) , but long-term studies are rare. 2. Timescale Separation : Node addition is assumed to happen much more slowly than edge addition. This assumption appears in most theoretical frameworks ( Holme and Saramäki 2019 ) but lacks empirical validation. 3. System Isolation : Networks are treated as isolated systems, ignoring external influences. Some recent studies suggest this may be an oversimplification ( Karimi et al. 2018 ) . 4. Equilibrium Dynamics : Many models assume the network reaches a quasi-steady state where structural properties stabilise ( Krioukov et al. 2016 ) . This may not reflect reality in rapidly growing networks. Empirical studies of network evolution over time have been limited in both duration and scale. Most cover periods shorter than a decade ( Mislove et al. 2008 ; Kumar et al. 2010 ) . Even recent studies of large online social networks typically analyse networks with millions of nodes over a few years ( Leskovec and Horvitz 2020 ) . Studies of network evolution over decades are rare, with notable exceptions being small citation networks ( Sinatra et al. 2015 ) and limited collaboration networks ( Vazquez et al. 2018 ) . Mathematical frameworks for analysing growing networks have advanced significantly. Recent developments include non-Markovian growth processes ( Williams et al. 2019 ) , temporal network motifs ( Paranjape et al. 2017 ) , and higher-order network structures ( Benson et al. 2018 ) . However, these frameworks typically assume the growth assumptions listed above. In this paper, we analyse two large collaboration networks over extended time periods: the Microsoft Academic Graph (MAG) covering 1800-2020, and the Internet Movie Database (IMDb) covering 1900-2020. These datasets provide an extensive historical perspective for studying network evolution: 1. Temporal span: >100 years of continuous evolution, 2. Scale: Hundreds of millions of nodes and edges,

Read the complete text