[Paper Review] Unraveling higher-order dynamics in collaboration networks

David Abella, Piero Birello+5 more|arXiv (Cornell University)|Jun 30, 2023

Complex Network Analysis TechniquesPhysics and Astronomy3 citations

TL;DR

This study introduces a temporal higher-order network analysis of scientific collaboration in the Geology domain using simplicial complexes to model multi-author publications. It reveals that group collaboration dynamics exhibit persistent aggregation and disaggregation patterns over time, with increasing variability in group size changes from 1980 to 2016, indicating evolving structural complexity in scientific teams beyond pairwise interactions.

ABSTRACT

The interactions between individuals play a pivotal role in shaping the structure and dynamics of social systems. Complex network models have proven invaluable in uncovering the underlying mechanisms that govern the formation and evolution of these systems. However, conventional network representations primarily emphasize pairwise interactions, represented as edges in the network. In reality, many social interactions occur within groups rather than individual pairs. To capture this crucial aspect, higher-order network representations come into play, especially to describe those complex systems that are inherently composed of agents interacting with group dynamics. Despite recent research advancements in exploring temporal higher-order networks in various systems, our understanding of collaboration networks remains limited. Specifically, there is a lack of knowledge regarding the patterns of group interactions within scientific collaborations. How do groups form and evolve in this context? In this study, we aim to delve into the temporal properties of groups within collaboration networks. Our investigation focuses on uncovering the mechanisms that govern the global, group, and individual-level dynamics, shedding light on how individuals collaborate and how groups form and disband over time. By studying these temporal patterns, we take a significant stride forward in comprehending the intricate dynamics of higher-order interactions within human collaboration systems.

Motivation & Objective

• To understand the temporal evolution of group collaborations in scientific networks beyond pairwise interactions.
• To identify mechanisms driving group formation, dissolution, and structural changes in multi-author scientific publications.
• To quantify the dynamics of collaboration size changes across consecutive years using higher-order network representations.
• To assess how collaboration group stability and variability have changed over time in the Geology domain.
• To provide a methodological framework applicable to other domains involving collective human collaboration.

Proposed method

• Models co-authorship data as a sequence of timestamped simplices, where each simplex represents a publication with its authors as nodes.
• Focuses on medium-sized simplices (5–20 authors) to analyze group-level dynamics and avoid noise from very small or large teams.
• Computes 'order difference' as the change in group size between consecutive years: positive for aggregation, negative for disaggregation.
• Stratifies order differences by year and by collaboration size to analyze temporal and size-dependent trends.
• Uses polynomial regression (order 3) to model the non-linear evolution of standard deviation in order differences over time.
• Performs regression analysis and time-window stratification (10-year intervals) to validate robustness of linear trends in average order differences.

Experimental results

Research questions

• RQ1 How do scientific collaboration groups evolve over time in terms of size and composition?
• RQ2 What are the dominant patterns of group aggregation and disaggregation in multi-author scientific publications?
• RQ3 How has the variability in group size changes evolved from 1980 to 2016 in the Geology domain?
• RQ4 What is the relationship between collaboration group size and the magnitude of size changes over time?
• RQ5 To what extent are observed dynamics in collaboration groups robust across different time windows and data subsets?

Key findings

• The standard deviation of order differences in collaboration group sizes increased non-linearly from 1980 to 2016, indicating growing dynamism and instability in team structures over time.
• A consistent linear relationship was observed between average order difference and collaboration size, with larger groups showing proportionally greater changes in membership over time.
• The trend of increasing variability in group size changes was robust across multiple 10-year time windows, confirming the stability of the observed temporal pattern.
• The analysis revealed that both aggregation and disaggregation are prevalent phenomena, with no dominant direction, suggesting fluid team composition in scientific collaborations.
• The method successfully captured higher-order temporal dynamics in collaboration networks, demonstrating its potential for studying collective human behavior beyond pairwise interactions.
• The dataset and code are publicly available, enabling reproducibility and extension of the framework to other scientific disciplines.