Source-linked AI summary
The Complex Network of Evolutionary Computation Authors: an Initial Study
Carlos Cotta, Juan J. Merelo
TL;DR
The paper asks what the evolutionary-computation co-authorship network reveals about disciplinary cohesion, collaboration, and author prominence. It analyzes DBLP data with automated network extraction and filtering, finding broadly familiar network properties alongside distinctive cohesion and proceeding-editorship effects. The study also identifies factors associated with sociometric prominence and argues that paper types should be analyzed separately.
Problem
The paper investigates the structure and cohesion of evolutionary-computation collaboration, including what makes some authors prominent and how proceeding editorship affects network properties.
Source: §1 · p. 2, §5 · p. 11Method
The authors build an EC co-authorship network from DBLP by traversing author pages, filtering EC papers, and seeding the search with active conference authors.
Source: §2 · p. 3Results
The EC network resembles computer-science co-authorship networks but shows closer author relationships, while proceeding co-editorship affects long-range links and centrality measures.
Source: §3 · p. 6, §4 · p. 10, §5 · p. 11Takeaways & Limitations
The study provides an initial account of EC collaboration structure and highlights proceeding editorship as an important source of distortion in network analyses.
Source: §5 · p. 11The authors suggest separating technical reports, conference papers, and journal papers because their different collaboration patterns may produce different networks.
Source: §5 · p. 11
Abstract
from arXiv · showhide
EC paper authors form a complex network of co-authorship which is, by itself, a example of an evolving system with its own rules, concept of fitness, and patterns of attachment. In this paper we explore the network of authors of evolutionary computation papers found in a major bibliographic database. We examine its macroscopic properties, and compare it with other co-authorship networks; the EC co-authorship network yields results in the same ballpark as other networks, but exhibits some distinctive patterns in terms of internal cohesion. We also try to find some hints on what makes an author a sociometric star. Finally, the role of proceeding editorship as the origin of long-range links in the co-authorship network is studied as well.
1 Introduction
This paper studies evolutionary-computation co-authorship as a complex network to characterize its structure, cohesion, central actors, and collaboration patterns.
1 Introduction
- Co-authorship networks connect authors who have written at least one paper together, enabling analysis of macroscopic structure and prominent actors.These networks can reveal patterns of collaboration and which authors or links stand out.Source: §1 · pp. 1–2
- Prior studies generally characterize co-authorship networks as small-world and scale-free, with preferential attachment helping explain why established authors gain more links.The paper positions its EC analysis within earlier work across physics, medicine, computer science, and other communities.Source: §1 · p. 2
- The study analyzes evolutionary-computation researchers’ co-authorship network to understand disciplinary cohesiveness, collaboration patterns, and factors associated with author prominence.It examines both network-level properties and the positions of individual authors and co-authorships.Source: §1 · p. 2
2 Materials and Methods
The authors construct the EC co-authorship network from DBLP using automated author-page traversal, publication filtering, and a broad seed set of active conference authors.
2 Materials and Methods
- The database may still contain name inconsistencies, including possible duplicate identities, although the authors report no glaring same-name duplicity in the EC community.This remains a data-quality consideration despite DBLP maintainers’ efforts to reduce naming problems.Source: §2 · p. 3
- DBLP supplies indexed author and publication data, moderated curation, regular page structure, and co-author hyperlinks that support automated network construction.Author-page URLs help identify authors, while moderated entries reduce submission-activity bias.Source: §2 · p. 3
- The scraping robot traverses author pages through co-author links, tracks processed authors to avoid cycles, and extracts papers and collaborators.Author keys are stored in a stack, and processed authors are tracked using an ordered binary tree.Source: §2 · p. 3
- EC papers are identified through conference or journal references and title keywords, with visual inspection finding false positives well below 1%.Most false positives were computational-biology papers and were removed from the network.Source: §2 · p. 4
- The seed set includes authors publishing during the previous five years in GECCO, PPSN, EuroGP, EvoCOP, and EvoWorkshops, expanding from 2,536 seeds to 5,492 network authors.CEC was not indexed in DBLP, but the authors state this did not alter macroscopic network properties.Source: §2 · p. 4
3 Macroscopic Network Properties
The EC co-authorship network broadly resembles other collaboration networks but has distinctive cohesion: a large connected component, high clustering, and short paths. Its productivity and collaboration distributions follow power-law patterns, while collaboration size reflects the role of proceedings.
Productivity
- EC authors average 2.9 papers, following Lotka’s Law with a long tail in which most authors appear once while some author dozens of papers.The EC sample is smaller than the broader comparison communities and resembles NCSTRL more closely in publication practices.Source: §3 · pp. 4–5
Distributions
- The number of papers per author follows a distribution with slope -2.00, while the number of authors per paper follows one with slope -5.27.These distributions summarize the heavy-tailed productivity and paper-size patterns observed in the EC network.Source: §3 · pp. 4–5, Figure 1 · p. 5
Collaboration patterns
- The average EC collaboration contains 2.56 authors, with a power-law distribution above three authors and a tail peak attributed to proceedings-based large collaborations.This average is below biomedical research and high-energy physics, similar to physics, and slightly above computer science.Source: §3 · pp. 5–6, Figure 1 · p. 5
- The EC network has 4.2 collaborators per author, below physics and biomedicine but above NCSTRL, suggesting openness to new collaborations from a computer-science perspective.The collaborator-count distribution fits a power law with exponent -2.58, consistent with preferential attachment but also constrained by practical collaboration limits.Source: §3 · p. 6
Giant component
- More than 2/3 of EC authors belong to the giant component, whose high clustering and intermediate mean distance indicate strong internal cohesion.The giant component is larger than in computer science but smaller than in physics or biomedicine; its mean distance lies between those communities and computer science.Source: §3 · p. 6, Figure 2 · p. 7
4 Evolutionary Computation Sociometric Stars
The study identifies sociometric stars in the EC co-authorship network using geodesic centrality measures, collaborator counts, and network-distance patterns. It also shows that proceedings strongly shape centrality rankings and create long-range links that reduce inter-author distances.
Centrality measures
- Closeness measures an actor’s access to information, whereas betweenness captures potential control over information flow through shortest paths.Shortest paths also form referral chains and sequences of related research topics between authors.Source: §4 · pp. 7–8
- Betweenness and closeness identify EC sociometric superstars, with betweenness showing larger top-rank gaps than closeness.Table 2 reports normalized betweenness and closeness values, with some actors appearing in both top-lists.Source: §4 · p. 8, Table 2 · p. 8
Collaborators and distance
- At least 23 collaborators correspond to mean distance below 4.61 for the top 5% of actors, while 33 collaborators are needed for statistical significance.The relationship is based on decreasing average distance as collaborator degree increases, with standard deviations shown in Figure 3.Source: §4 · p. 9, Figure 3 · p. 9
Proceedings and network structure
- The proceedings-free network has a more diffuse core, plausibly reflecting several micro-cores associated with different EC subareas.Figure 4 compares the network structure and author-distance distributions before and after proceedings removal.Source: §4 · p. 9, Figure 4 · p. 10
- Proceedings create long-range links because editors often span thematic areas, lowering average inter-actor distances across the network.Without proceedings, average and maximum distances increase by 2 units, modal distance by 3 units, and the distribution becomes more symmetric.Source: §4 · p. 10, Figure 4 · p. 10
Proceedings and centrality
- Proceedings removal makes centrality rankings more concordant and elevates previously underrepresented researchers, including Z. Michalewicz.After removal, 7/10 top positions agree across the two centrality measures, compared with 6/10 before; Michalewicz ranks highest in closeness, fifth in betweenness, and seventh in collaborators.Source: §4 · p. 10, Table 3 · p. 11
5 Discussion and Conclusion
The preliminary EC co-authorship study finds a network broadly similar to computer science but with closer author relationships, while proceedings co-editorship substantially affects network structure and centrality. It recommends separating paper types in future analyses and extending the work to temporal evolution, funded networks, and EC communities.
5 Discussion and Conclusion
- Proceedings co-editorship substantially distorts the EC network’s overall structure and centrality measures, motivating its exclusion from future studies.The authors report that removing proceedings-based links changes network features dramatically and alters the network’s visual core.Source: §5 · p. 11
- The EC co-authorship network resembles computer science overall, yet its authors are more closely related to each other.Source: §5 · p. 11
- Different paper types may encode different collaboration relationships, so technical reports, conference papers, and journal papers should be analyzed separately.The paper characterizes technical reports as preliminary, conference papers as longer-term, and journal papers as reflecting more committed relationships.Source: §5 · p. 11
- Future work will examine network evolution, funded and transnational grants, and invisible colleges or communities within evolutionary computation.Source: §5 · p. 11