- The paper presents a novel citation-based clustering method that leverages direct citation relations and a smart local moving algorithm to optimize publication grouping.
- It applies CitNetExplorer for detailed publication-level network analysis and VOSviewer for aggregate visualizations and thematic mapping.
- The study analyzes over 100,000 astronomy publications, revealing both intricate internal structures and broad thematic clusters to advance bibliometric research.
Citation-Based Clustering of Publications Using CitNetExplorer and VOSviewer
In the paper "Citation-based clustering of publications using CitNetExplorer and VOSviewer," Nees Jan van Eck and Ludo Waltman, from the Centre for Science and Technology Studies at Leiden University, present a methodology for clustering scientific publications utilizing citation relations. The approach leverages two software tools: CitNetExplorer and VOSviewer. This paper offers a comprehensive demonstration of these tools' applications using a large dataset from the field of astronomy and astrophysics.
Methodological Overview
The central thrust of this paper lies in using direct citation relations to cluster publications. CitNetExplorer and VOSviewer serve distinct but complementary purposes in the analysis of these clusters:
- CitNetExplorer: Primarily focuses on publication-level analysis. It employs clustering techniques grounded in network science to organize publications based on direct citation relations. CitNetExplorer provides functionalities for visualizing citation networks, allows users to drill down into specific clusters for more detailed analysis, and includes search capabilities.
- VOSviewer: This tool is used to analyze clustering results at an aggregate level. It provides visual representations of clusters and inter-cluster citation relations and constructs term maps to depict the thematic topics covered by publications within a cluster.
Clustering Technique
CitNetExplorer's clustering technique calculates the relatedness of publications purely based on direct citation relations as opposed to bibliographic coupling or co-citation relations. The authors argue that direct citation provides a more accurate representation of relatedness and mitigates computational challenges associated with larger datasets.
Publications are assigned to clusters by optimizing a quality function similar to the modularity function developed by Newman and Girvan, but adjusted to avoid the well-known resolution limit problem. The clustering technique is supported by the smart local moving algorithm, which generally yields higher quality clustering solutions compared to the Louvain algorithm.
Demonstration and Results
To exhibit the practical utility of CitNetExplorer and VOSviewer, the authors used a large dataset consisting of over 100,000 publications in astronomy and astrophysics. The clustering process included:
- Initial Clustering with CitNetExplorer: Four clustering solutions were created at different levels of detail. The number of clusters varied from 22 clusters at the highest level of aggregation to 434 clusters at the most detailed level.
- Detailed analysis of clusters: Cluster properties were examined through CitNetExplorer, illustrating how the top 100 most frequently cited publications were visualized. This revealed the internal structures and citation relationships among clusters.
- Aggregate analysis using VOSviewer: Provided visual summaries of clusters and their interdependencies. For example, one visualization illustrated 22 main clusters, organized into four major groups, giving insight into broad thematic areas within the dataset. Additionally, term maps helped characterize the topics covered by specific clusters, such as solar phenomena, providing deep insights into the thematic structure of the clustered publications.
Implications and Future Work
This paper contributes substantially to the toolbox of bibliometricians by highlighting how sophisticated cluster analyses can be carried out without extensive expertise in clustering algorithms or advanced computing skills. The freely available CitNetExplorer and VOSviewer tools enable detailed and comprehensive analysis of publication networks and are particularly useful for those looking to understand thematic clusters within large bibliometric datasets.
However, the authors acknowledge the somewhat laborious nature of using the two tools in conjunction and recognize limitations such as the inability to assign some publications to clusters due to the lack of direct citation relations. They suggest future work on an integrated tool combining the strengths of both CitNetExplorer and VOSviewer, potentially enhancing the ease of use and functionality.
The integration aims to support users through different types of visualizations, assist exploration of citation relations both at macro and micro levels, and reveal temporal dynamics in interest within scientific fields.
Conclusion
Nees Jan van Eck and Ludo Waltman's work on citation-based clustering using CitNetExplorer and VOSviewer offers an effective and user-friendly methodology for analyzing large-scale bibliographic data. Their contributions lie not only in the development of these tools but also in elucidating complex citation networks, thus advancing the capacity of bibliometric research. Further developments in integrating these tools could streamline bibliometric analysis processes, providing even richer insights into the structure and evolution of scientific knowledge.