Source-linked AI summary
Discovering author impact: A PageRank perspective
Erjia Yan, Ying Ding
TL;DR
Citation counts do not fully capture author impact, motivating a network-based complement. The paper proposes citation-weighted PageRank for coauthorship networks, tests damping factors, and compares the measure with established indicators. The authors report reliable author-impact evaluation, including strong agreement across damping factors and favorable comparisons with h-index and standard PageRank.
Problem
Citation counts may not provide a complete perspective on author impact, motivating a complementary coauthorship-network measure.
Method
The paper proposes weighted PageRank that combines citation information with coauthorship-network topology and evaluates it under different damping factors.
Results
The weighted PageRank is reported as reliable, accurately identifying Price Award winners near the top of rankings and outperforming h-index and PageRank.
Takeaways & Limitations
Weighted PageRank offers an alternative measure that integrates authors’ academic and community impact for informetrics author evaluation.
Takeaways & Limitations
The analysis uses the largest coauthorship component, containing 1,034 authors, and citation and h-index comparisons are subject to this study’s data.
Abstract
from arXiv · showhide
This article provides an alternative perspective for measuring author impact by applying PageRank algorithm to a coauthorship network. A weighted PageRank algorithm considering citation and coauthorship network topology is proposed. We test this algorithm under different damping factors by evaluating author impact in the informetrics research community. In addition, we also compare this weighted PageRank with the h-index, citation, and program committee (PC) membership of the International Society for Scientometrics and Informetrics (ISSI) conferences. Findings show that this weighted PageRank algorithm provides reliable results in measuring author impact.
1 Introduction
Citation counts indicate publication impact but may not fully capture author impact. The paper therefore uses coauthorship-network topology and PageRank as complementary perspectives, applying a weighted approach in informetrics.
- Citation counts are commonly used to measure scientific publication impact, but they may not provide a complete view of author impact.
- Coauthorship analysis complements citation analysis by capturing individual actors together with the topology of their collaboration network.
- PageRank weights authors connected to diverse collaborators and to highly coauthored authors, offering a complementary perspective on author impact.
- The study evaluates informetrics authors using correlations, damping factors from 0.15 to 0.85, weighted PageRank rankings, and comparisons with h-index, citations, and ISSI conference PC membership.
2 Related studies
Related studies apply PageRank and network centrality to scientific influence, journals, countries, publications, and coauthorship. Prior coauthorship work suggests PageRank-based measures can align with expert or program-committee judgments.
- Collaboration patterns vary by discipline, so disciplinarity should be considered when comparing coauthorship networks.
- PageRank has been applied to publication, journal, country, and coauthorship networks to assess research output or influence.
- Weighted PageRank variants assign different importance to citations or coauthor ties based on network structure or collaboration intensity.
- In digital-library coauthorship, PageRank and AuthorRank matched Joint Conference on Digital Libraries program committee members more precisely than centrality measures.
3 Methodology
The methodology represents authors and coauthor relations as a weighted network, applies PageRank and citation-weighted personalization, and tests damping-factor effects. The study uses informetrics data and focuses the main PageRank analysis on the largest connected component.
- Data and network: The dataset contains 6,049 authors and 7,358 coauthor ties retrieved from Web of Science informetrics-related records.
- PageRank: PageRank ranks graph vertices using inbound and outbound links, with damping factor d controlling link-following probability and 1-d representing random jumps.
- Data and network: In the coauthorship network, nodes represent authors, edges represent coauthor relations, and edge weights represent coauthor frequency.
- Weighted PageRank: The weighted PageRank extends PageRank by incorporating citation counts with coauthorship-network topology into the personalization component.
4 Results and analysis
The results examine how PageRank relates to citation rankings, how damping factors affect rankings, and whether weighted PageRank aligns with author-impact indicators. Weighted PageRank combines citation and coauthorship topology, while its ranking behavior varies with damping-factor emphasis.
- Correlation between PageRank and citation rank: 0.4673 was the overall correlation between citation and PageRank, with the highest correlation occurring among the top 100 authors.Correlation declined from 0.4673 to 0.0540 toward the back of the ranking; the authors argue that only the top 10%-20% provide useful PageRank data.
- Correlation between PageRank and citation rank: Lower citation values covered a wide range of PageRank values, producing horizontal rather than diagonal patterns in the scatter plot.More nodes were distributed around the diagonal at higher values, indicating stronger correspondence there.
- PageRank with different damping factors: PR scores under different damping factors had pairwise correlations above 0.95, with adjacent factors exceeding 0.99 and the remote pair 0.85 versus 0.15 reaching 0.95.Changes were concentrated mainly in the middle of the ranking, while the top and tail converged toward common distribution lines.
- Evaluation: PR_W combines citation-based academic impact with coauthorship-based community impact and was compared with h-index, citation, and PC membership measures.At d=0.55, citation and h-index matched PR_W more precisely than PR and PC membership, while PR_W occupied a middle position between academic- and community-impact measures.
- Weighted PageRank: For d=0.85, coauthorship-network topology had greater influence, whereas for d=0.15, citation counts had greater influence.Authors such as Rousseau and Kretschmer ranked higher at d=0.85, while highly cited authors such as White and Small ranked higher at d=0.15.
5 Conclusion
The study extends citation-based author evaluation with PageRank on coauthorship networks and develops a weighted version integrating citation and network topology. The weighted algorithm provides an alternative and reliable measure of author impact, with little global reordering across damping factors.
- PageRank provides a meaningful extension to traditional citation counts for evaluating authors.
- PR and citation are correlated, but PR may be most useful for the top 10% to 20% of authors.The authors relate this pattern to power-law distributions and limited regular participation in scientific collaboration.
- Correlation coefficients between PR values under different damping factors are all above 0.95.The authors conclude that damping factors have little influence on PR values in coauthorship networks.
- PR_W integrates citation and coauthorship topology to measure academic and community impact.The weighted algorithm accurately identifies Price Award winners near the top of the ranking and outperforms h-index and PR in that comparison.