Source-linked AI summary
Citation-based clustering of publications using CitNetExplorer and VOSviewer
Nees Jan van Eck, Ludo Waltman
TL;DR
Making sense of clustering solutions can be a serious challenge, especially when clustering techniques are not straightforward. The paper demonstrates using CitNetExplorer and VOSviewer to cluster publications based on direct citation relations and analyze the resulting solutions, including more than 100,000 astronomy and astrophysics publications. The tools support sophisticated cluster analyses without data preprocessing, although publications cannot always be properly assigned to a cluster.
Problem
Making sense of clustering solutions can be a serious challenge, especially when clustering techniques are not straightforward.
Method
CitNetExplorer and VOSviewer are used to cluster publications based on direct citation relations and analyze the resulting clustering solutions.
Results
More than 100,000 publications in astronomy and astrophysics were clustered using CitNetExplorer.
Takeaways & Limitations
The demonstrated tools support sophisticated cluster analyses without data preprocessing.
Takeaways & Limitations
Publications cannot always be properly assigned to a cluster, and the tools' ease of use does not eliminate all methodological challenges.
Abstract
from arXiv · showhide
Clustering scientific publications in an important problem in bibliometric research. We demonstrate how two software tools, CitNetExplorer and VOSviewer, can be used to cluster publications and to analyze the resulting clustering solutions. CitNetExplorer is used to cluster a large set of publications in the field of astronomy and astrophysics. The publications are clustered based on direct citation relations. CitNetExplorer and VOSviewer are used together to analyze the resulting clustering solutions. Both tools use visualizations to support the analysis of the clustering solutions, with CitNetExplorer focusing on the analysis at the level of individual publications and VOSviewer focusing on the analysis at an aggregate level. The demonstration provided in this paper shows how a clustering of publications can be created and analyzed using freely available software tools. Using the approach presented in this paper, bibliometricians are able to carry out sophisticated cluster analyses without the need to have a deep knowledge of clustering techniques and without requiring advanced computer skills.
1. Introduction
The paper addresses the challenge of interpreting very large publication clustering solutions by demonstrating how CitNetExplorer and VOSviewer can create and analyze them. CitNetExplorer supports publication-level analysis, while VOSviewer complements it with aggregate-level visualizations.
- Motivation: Large publication clustering solutions can contain thousands or millions of publications, making them difficult to interpret.The difficulty is especially pronounced when clustering is applied to individual publications.
- Contribution: The paper demonstrates how CitNetExplorer and VOSviewer can cluster publications and analyze the resulting solutions.Both tools are freely available software tools used for bibliometric cluster analysis.
- CitNetExplorer: CitNetExplorer clusters publications based on citation relations and analyzes clustering solutions at the level of individual publications.Its approach includes visualizing citation networks and drilling down into selected clusters.
- VOSviewer: VOSviewer analyzes CitNetExplorer clustering solutions at an aggregate level using cluster and term-map visualizations.These visualizations show relations between clusters and the topics covered by them through important terms and their co-occurrences.
- Demonstration: The demonstration applies CitNetExplorer to more than 100,000 astronomy and astrophysics publications and uses both tools to analyze the resulting clusters.The tools rely strongly on visualizations to facilitate analysis.
2. Clustering technique
The clustering technique determines publication relatedness from direct citation relations and assigns publications to clusters by maximizing a quality function. Its design avoids the modularity resolution limit but has important scope and parameter limitations.
- Determining relatedness: The technique determines publication relatedness from direct citation relations rather than word relations, bibliographic coupling, or co-citation relations.The authors prefer direct citations because alternative relations can be less accurate, ambiguous, or computationally demanding.
- Determining relatedness: Direct citation relations cannot properly assign publications with no direct citation links, especially when the analysis period is short.This is an author-acknowledged limitation of the chosen relatedness measure.
- Assigning publications: The method assigns each publication to exactly one cluster by maximizing a quality function based on publication relatedness.The approach therefore produces non-overlapping clusters with no unassigned publications in the formal assignment scheme.
- Quality function: The quality function does not suffer from the resolution limit problem affecting modularity.The authors state that modularity can yield counterintuitive results in certain situations.
- Resolution parameter: Higher values of the resolution parameter γ produce more clusters, but no generally optimal γ exists.The authors recommend trying different values and choosing one that provides useful results for the specific analysis.
- Optimization and cluster structure: The smart local moving algorithm is used to maximize the quality function and typically finds higher-quality clustering solutions than Louvain with similar computing time.The method usually yields a limited number of larger clusters and more smaller clusters; small clusters may be discarded or merged.
3. Results
The demonstration applies CitNetExplorer and VOSviewer to cluster a large astronomy and astrophysics publication set and inspect the resulting solutions at publication and aggregate levels. The analyses produce multiple levels of detail and reveal broad literature structure, while the authors do not assess cluster quality or compare alternative solutions.
- Scope and limitation: The authors did not assess clustering quality or compare their solutions with alternatives because they lacked the domain knowledge for in-depth cluster interpretation and quality assessment.They refer readers to another paper for comparison with alternative solutions.
- Data and preprocessing: 111,616 publications and 929,364 within-set citation relations formed the astronomy and astrophysics data set.CitNetExplorer removed 3,824 relations, leaving 925,540 citation relations for analysis.
- Clustering solutions: Four clustering solutions varied in detail through different resolution and minimum cluster-size parameters.Clusters below the minimum size were merged with larger clusters.
- Clustering solutions: Level 1 contained 22 clusters averaging 4,629 publications, whereas level 4 contained 434 clusters averaging 235 publications.The largest cluster was almost 15,000 publications at level 1 and somewhat more than 1,000 at level 4.
- Clustering solutions: Publication distributions across clusters were quite skewed regardless of the clustering solution’s level of detail.The authors describe this skew as typical when their clustering technique is used.
- Visualization and analysis: CitNetExplorer visualized individual publications and citation relations, while VOSviewer summarized clusters and their topics at an aggregate level.CitNetExplorer visualizations encode time vertically and citation relatedness horizontally; VOSviewer term maps use titles and abstracts.
- Visualization and analysis: The level 1 visualization showed relatively independent literature bodies, with most citation relations occurring within clusters, while VOSviewer indicated broad astronomy subfields.The broad interpretations included astroparticle physics, gravitational physics, cosmology, galaxies, stars, solar physics, and planetary science.
4. Conclusion
The paper demonstrates how CitNetExplorer and VOSviewer can cluster publications and analyze the resulting solutions through complementary visualizations. The approach is freely available and accessible, but requires basic clustering knowledge and has limitations in direct-citation coverage and workflow integration.
- Conclusion: CitNetExplorer and VOSviewer support clustering publications and analyzing the resulting clustering solutions.The tools are freely available for bibliometric analysis.
- Conclusion: CitNetExplorer focuses on individual publications, whereas VOSviewer analyzes clustering solutions at an aggregate level.Their visualizations are intended to complement one another.
- Conclusion: Meaningful analysis still requires a basic understanding of clustering techniques to avoid misinterpreting results.The tools do not require advanced computer skills, and Web of Science downloads can be used directly as input without preprocessing.
- Conclusion: Direct citation relations allow efficient clustering of very large publication sets, potentially reaching tens of millions of publications.Some publications cannot be properly assigned to a cluster when they lack direct citation relations.
- Conclusion: The combined workflow is somewhat laborious because preparing VOSviewer input from CitNetExplorer results is not entirely straightforward.The authors plan an integrated tool combining key features of both applications.
- Conclusion: The planned integrated tool would support interactive exploration of hierarchical clusters, citation relations, cluster topics, and topic dynamics over time.It would combine different types of interactive visualizations for exploring scientific literature.
Appendix
The appendix contains an extended summary of 22 level 1 clusters and associated astronomy, astrophysics, physics, and geophysics terms, journals, and publications.
- Appendix: Table A1 is an extended summary of the 22 level 1 clusters.The supplied appendix material identifies the table but does not provide further table-level findings.
- Appendix: The appendix associates clusters with journals and named publications from astronomy, astrophysics, physics, and related fields.Listed sources include Astronomy & Astrophysics, Astrophysical Journal, Physical Review D, and Icarus.