Source-linked AI summary
Betweenness Centrality as a Driver of Preferential Attachment in the Evolution of Research Collaboration Networks
Alireza Abbasi, Liaquat Hossain, Loet Leydesdorff
TL;DR
The paper asks whether preferential attachment in scientific coauthorship networks differs according to authors’ forms of centrality. Using a longitudinal steel-structures coauthorship database, it compares degree, closeness, and betweenness centrality. It finds that betweenness, particularly as brokering, is more strongly associated with preferential attachment than degree or closeness.
Problem
The paper asks whether preferential attachment in scientific collaboration networks depends on existing authors’ network positions and which positions attract new authors.
Method
The study analyzes a coauthorship network in the steel-structures research field using degree, closeness, and betweenness centrality measures.
Results
Betweenness centrality and authors’ brokering position are more important for preferential attachment than degree or closeness centrality.
Takeaways & Limitations
Preferential attachment in the studied coauthorship network is associated with authors’ brokering position rather than only their degree.
Takeaways & Limitations
The study examines only one field, steel structures, so its contribution is mainly a hypothesis.
Abstract
from arXiv · showhide
We analyze whether preferential attachment in scientific coauthorship networks is different for authors with different forms of centrality. Using a complete database for the scientific specialty of research about "steel structures," we show that betweenness centrality of an existing node is a significantly better predictor of preferential attachment by new entrants than degree or closeness centrality. During the growth of a network, preferential attachment shifts from (local) degree centrality to betweenness centrality as a global measure. An interpretation is that supervisors of PhD projects and postdocs broker between new entrants and the already existing network, and thus become focal to preferential attachment. Because of this mediation, scholarly networks can be expected to develop differently from networks which are predicated on preferential attachment to nodes with high degree centrality.
1 INTRODUCTION
The paper examines how authors’ network positions relate to preferential attachment during the evolution of scientific coauthorship networks. It asks whether degree, closeness, or betweenness positions are most attractive to new authors.
- Research problem: The paper investigates whether preferential attachment applies to social collaboration networks, where its applicability remains debated.It frames this question against models in which new nodes preferentially attach to well-connected existing nodes.
- Motivation: Scientific collaboration networks grow through changing author connections, making attachment behavior relevant to their structural evolution.The study focuses on how authors’ positional properties relate to structural change and evolutionary behavior.
- Approach: Using a coauthorship network in research about steel structures, the study compares authors’ degree, closeness, and betweenness centrality.These measures represent being well connected, close to others, or able to broker between authors.
- Attachment patterns: The paper considers four attachment types, including links between new authors and existing authors and new links among previously unconnected existing authors.It distinguishes these from links among new authors and repeated links among already collaborating existing authors.
- Research questions: The study asks whether existing authors’ positions associate with the number of new authors collaborating with them at a later time.It also asks which types of positions are most attractive for preferential attachment.
2 SOCIAL NETWORK ANALYSIS AND PREFERENTIAL ATTACHMENT
This section introduces social network analysis and preferential attachment as frameworks for studying how network structures form and evolve. In coauthorship networks, authors are nodes and coauthorships are links, while preferential attachment favors existing nodes with more links.
- Social network analysis: Social network analysis studies nodes and relations in networks that can represent individuals, groups, organizations, or countries.Its applications include social influence, groupings, inequality, disease propagation, information communication, and organizational analysis.
- Coauthorship networks: In scientific collaboration networks, authors are nodes and coauthorship relations are ties, with a tie indicating at least one joint publication.Link weights can represent the number of publications jointly produced by two authors.
- Network evolution: Evolving-network models commonly combine growth with preferential attachment.Growth adds nodes and links, while preferential attachment specifies how new nodes select existing ones.
- Preferential attachment: Preferential attachment states that new nodes favor existing nodes that are already well connected, operationalized through degree centrality.This mechanism is associated with scale-free network patterns and cumulative advantage.
- Preferential attachment: Preferential attachment models aim to specify mechanisms generating observed network structures rather than merely describing network topology.The process is also described as the rich-get-richer principle, cumulative advantage, or the Matthew effect.
3 DATA AND MEASURES
The study constructs a longitudinal coauthorship dataset for the steel-structures field and analyzes author positions using social-network centrality measures. After cleaning, the database contains 1,869 publications by 3,004 authors across 1,324 institutes and 77 countries.
- Data preparation: The database required manual checks and merging of differently named universities and departments to complete affiliation information.The original affiliation data contained missing fields and inconsistent spellings.
- Data sample: The dataset covers publications on steel structures from 1999 through 2009, identified using the phrase “steel structure” in selected publication fields.The search used titles, keywords, or abstracts in the field’s top 15 journals and restricted results to English publications.
- Measures: The analysis uses centrality measures to evaluate authors’ strategic positions and their association with expected new coauthorships.The measures are used to study how network positions relate to attachment behavior.
Degree Centrality
Degree centrality measures a node’s local connectedness by counting its direct connections. In this paper, it represents an author’s popularity and activity through knowing more collaborators.
- Degree Centrality: Degree centrality counts the number of other nodes directly connected to a node.It is treated as a local centrality measure.
- Degree Centrality: The degree centrality of node k is defined from its connections to the other nodes in a network.The formulation uses a binary connection indicator between nodes i and k.
- Degree Centrality: High degree centrality reflects an author’s popularity and activity through knowing more people.Highly connected nodes may also function as informal group leaders.
Closeness Centrality
Closeness centrality measures a node’s global proximity to others using shortest-path distances. It is also associated with efficient information access and dissemination, but is problematic in unconnected networks.
- Closeness Centrality: Closeness centrality treats a node as globally central when it lies at short distances from many other nodes.The concept originated as a sum of geodesic distances and uses its inverse, or reciprocal distances, to represent closeness.
- Closeness Centrality: In unconnected networks, every node is infinitely distant from at least one other node, creating a problem for closeness centrality.The standard distance-based calculation therefore encounters a structural limitation when the network is disconnected.
- Closeness Centrality: Freeman’s formulation defines closeness as the sum of a node’s reciprocal distances to all other nodes.For node k, the distances d(pi, pk) are geodesic, or shortest-path, distances.
- Closeness Centrality: A node close to all others can obtain information efficiently and disseminate it quickly through the network.Closeness centrality is therefore described as a proxy for communication independence and efficiency.
Betweenness Centrality
Betweenness centrality is a global measure based on how often a node lies on shortest paths between other nodes. Such nodes can broker connections, control information flows, and act as gatekeepers within networks.
- Betweenness Centrality: The measure counts how often a particular node lies between other nodes in the network.Its formulation uses geodesic distances between nodes and geodesic paths that pass through the focal node.
- Betweenness Centrality: Betweenness centrality measures the proportion of shortest paths between node pairs that pass through a given node.For node k, the denominator counts shortest paths between pairs regardless of whether they pass through k.
- Betweenness Centrality: Nodes with high betweenness centrality act as brokers or gatekeepers connecting nodes and subgroups.Their structural position places them between otherwise separated parts of the network.
- Betweenness Centrality: Because other nodes depend on them, highly betweenness-central nodes can control information flows and indicate power or influence.The passage links brokerage position to both network communication and actors’ standing in groups or organizations.
4 ANALYSIS AND RESULTS
The steel-structures coauthorship network expanded substantially from 1999 to 2009, while new links increasingly connected authors within the evolving network. Attachment analyses show that betweenness centrality was more strongly associated with attracting new authors than degree or closeness centrality, with its importance increasing over time.
- Network growth: The network grew from 229 authors and 234 links in 1999 to 409 authors and 788 links in 2009.Authors almost doubled, while links increased more than three times.
- Network growth: 1.02 to 1.96 average links per author accompanied an increasing trend in collaborations between 1999 and 2009.The yearly increase fluctuated, but the overall average nearly doubled.
- Attachment behaviour: Most new links occurred among newly added authors, while relatively few new authors attached to existing authors.The 2000 network likewise showed few links between new and existing authors compared with links within the new or existing groups.
- Preferential attachment: The correlation between existing authors’ betweenness centrality and next-year attachment frequency was always significant and higher than correlations for degree or closeness centrality.Attachment frequency counted the number of new authors and links attached to existing authors.
- Preferential attachment: As the network grew, the correlation with betweenness centrality increased, whereas the degree correlation remained almost constant and closeness fluctuated.The results associate network evolution with increasing importance of betweenness centrality for attachment.
- Preferential attachment: Authors with high betweenness centrality had the largest average number of new co-authors in each period, producing a gap relative to high-degree or high-closeness authors.The comparison also found that high-centrality authors generally attracted more new co-authors than low-centrality authors.
5 DISCUSSION AND CONCLUSION
In the steel-structures coauthorship network, all three centrality measures correlate with new authors’ attachment, but betweenness centrality is the strongest predictor. The findings suggest that brokering authors, often supervisors connecting students with established colleagues, become focal points for preferential attachment, although the evidence is limited to one scientific field.
- Study and research question: Between 1999 and 2009, the study assessed whether degree, closeness, and betweenness centrality predicted new co-authorships in steel-structures research.The analysis examined the temporal evolution of co-authorship relations and the expectation of new co-authorships.
- Centrality and preferential attachment: All three centrality measures of existing authors correlate with the attachment frequency of new authors, but betweenness predicts attachment better than degree or closeness.Authors with greater betweenness attract more new co-authors than those with more co-authorship links or more direct connections to other nodes.
- Network evolution: A relatively small number of new authors attach to existing authors, while only about 8% of existing authors form a new collaboration in the following year.Existing authors’ new-collaboration rate is described as almost equal to that of new authors.
- Interpretation: Existing authors rarely initiate collaborations with previously published domain authors and instead prefer new authors, possibly reflecting supervisor–student collaborations.Authors with high betweenness appear to be supervisors who connect students with colleagues or other students through their publication histories.
- Centrality and preferential attachment: During network evolution, betweenness centrality indicates preferential attachment more strongly than the number of accumulated links, or degree centrality.This identifies a brokering position, rather than simply a larger local neighborhood, as the stronger correlate of new entrants’ attachments.
- Scope and limitation: Because the study analyzes only steel-structures research, its contribution is mainly a hypothesis requiring investigation across other scientific domains.The authors state that generalization requires examining additional fields.