Source-linked AI summary
Applying centrality measures to impact analysis: A coauthorship network analysis
Erjia Yan, Ying Ding
TL;DR
Most coauthorship-network research emphasizes macro-level topology, leaving individual actors’ positions and opportunities less examined. This study analyzes an evolving LIS coauthorship network with four centrality measures and finds that all are significantly correlated with citation counts, while also revealing different author-career trajectories.
Problem
Prior coauthorship research largely emphasizes overall network topology, with less attention to individual network properties and the relationship between centrality and citations.
Method
The study constructs an evolving coauthorship network from 16 leading LIS journals across 1988–2007 and calculates closeness, betweenness, degree, and PageRank centrality for authors.
Results
All four centrality measures are significantly correlated with citation counts, while the evolving network distinguishes consistently highly ranked, rising, and fading authors.
Takeaways & Limitations
Centrality measures can support impact analysis by capturing author collaboration scope, network position, and field impact alongside citation counts.
Takeaways & Limitations
Current centrality algorithms can overstate impact from multi-author papers and treat network position as a proxy rather than a direct measure of academic impact.
Abstract
from arXiv · showhide
Many studies on coauthorship networks focus on network topology and network statistical mechanics. This article takes a different approach by studying micro-level network properties, with the aim to apply centrality measures to impact analysis. Using coauthorship data from 16 journals in the field of library and information science (LIS) with a time span of twenty years (1988-2007), we construct an evolving coauthorship network and calculate four centrality measures (closeness, betweenness, degree and PageRank) for authors in this network. We find out that the four centrality measures are significantly correlated with citation counts. We also discuss the usability of centrality measures in author ranking, and suggest that centrality measures can be useful indicators for impact analysis.
1. Introduction
Coauthorship networks have largely been studied through macro-level topology and dynamics, while this article turns to micro-level properties such as individual power, ranking, and inequality.
- Recent social-network research increasingly examines networks with thousands or millions of vertices, emphasizing topology and dynamics.
- Coauthorship networks are an important form of social network studied within this large-scale research movement.
- Macro-level analyses describe overall structure, whereas micro-level analyses examine individual actors’ constraints, opportunities, power, stratification, ranking, and inequality.
2. Backgrounds
Centrality measures developed in sociology have been applied to power, prestige, journal impact, and coauthorship analysis, including comparisons with external rankings.
- Freeman’s work developed centrality concepts that later became degree, closeness, betweenness, and eigenvector centrality.
- Weighted PageRank and betweenness centrality have been applied to journal impact analysis, producing rankings with both overlaps and differences from impact-factor rankings.
- Coauthorship studies have applied centrality measures to digital-library and conference research communities, including comparisons with program-committee rankings.
- In one coauthorship study, betweenness centrality performed best among three compared centrality measures.
3. Methodology
The study applies four centrality measures to an evolving LIS coauthorship network built from twenty years of data across 16 journals. The measures capture different network positions, collaboration connections, and influence pathways.
- Centrality Measures: The study applies degree, closeness, betweenness, and PageRank to the coauthorship network.
- Centrality Measures: Degree centrality counts a vertex’s ties, although many ties may arise from coauthoring one paper with many authors.
- Centrality Measures: PageRank assigns rank through propagated backlink weights and is described here as directed weighted degree centrality.
- Centrality Measures: Closeness centrality emphasizes a vertex’s distances to all others and represents how quickly information can spread across the network.
- Centrality Measures: Betweenness centrality counts shortest paths through a vertex, identifying vertices that connect different groups.
- Data processing: The dataset covers 16 leading LIS journals from 1988–2007 and includes 10,344 articles and review articles after excluding anonymous articles.
4. Results and analysis
The LIS coauthorship network contains 10,579 authors and becomes more collaborative over time, while centrality distributions reveal highly unequal author positions. Centrality rankings generally align with citation rankings, but discrepancies arise for authors with limited publication coverage or changing collaboration patterns.
- Network overview: 10,579 authors formed the LIS coauthorship network, with averages of 2.40 papers per author, 1.80 authors per paper, and 2.24 collaborators per author.These averages were lower than corresponding values reported for biology and physics coauthorship networks.
- Network evolution: The mean distance increased from 2.49 in 1992 to 9.68 in 2007 as new authors joined with limited collaboration scope.The paper contrasts this with a decrease in neuro-science coauthorship distance from 10 in 1991 to 6 in 1998.
- Centrality distributions: Betweenness, degree, and PageRank follow power-law distributions, whereas closeness follows a normal distribution.The degree distribution is described as consistent with a scale-free network.
- Author ranking over time: A few authors remained highly ranked across all four time slices, including Willett, P, while others rose or declined with changing publication and collaboration patterns.The paper gives Willett, P as a persistent high-ranking example and Thelwall, M as a newer high-ranking author.
- Citations and centrality: The seven most cited authors had very low centrality rankings because they had few papers, few collaborators, and peripheral network positions.Their highly cited publications were therefore not matched by strong structural centrality in this dataset.
- Citations and centrality: Citation-count rankings broadly matched centrality rankings, although incomplete publication coverage affected some authors’ ranking results.The overall ranking alignment is shown using PageRank as the benchmark in the ranking-distribution comparison.
5. Discussion and Conclusion
The evolving coauthorship network reveals changing author collaboration positions and career trajectories, while centrality measures correlate significantly with citation counts. Centrality can integrate article impact with author field impact, but current algorithms have important limitations that motivate further refinement.
- Dynamic author positions: Authors’ centrality rankings across time reveal stable, rising, and fading career trajectories.Some authors remain highly ranked throughout all periods, while others rise or fade from the field.
- Centrality and impact: All four centrality measures are significantly correlated with citation counts, although inconsistencies occur.The measures capture both article impact and aspects of an author’s position or field impact.
- Centrality and impact: Centrality combines article impact measured by citations with author field impact associated with social capital.This broader scope distinguishes centrality from citation counts alone.
- Algorithmic limitations: Current centrality algorithms can misrepresent academic impact through coauthor counts, network proximity, and interdisciplinary brokerage.A ten-author paper gives each author degree centrality of 9, equivalent to 45 two-author papers under the described calculation; closeness and betweenness can likewise be inflated by collaborators or interdisciplinary ties.
- Future directions: Improved centrality algorithms and broader network or semantic information are proposed for stronger impact evaluation.Suggested directions include refined measures, applications to other social networks, and semantic ontologies for systematic evaluation indicators.