ClusterChoice: a web-based tool for interpretable clustering and algorithm comparison in bibliometric networks

Cluster analysis is widely used in bibliometric research, but dense networks often reduce readability, and clustering algorithms or quality indicators are frequently underreported. This study introduces ClusterChoice , a web-based R Shiny application for selecting interpretable clustering structures in bibliometric networks. The software accepts co-word occurrence, document–term matrix, and node–edge data, applies top-node filtering to reduce computational burden, and compares 11 clustering algorithms using modularity Q, mean silhouette score, ARI, and NMI. A SoftwareX bibliographic dataset was used to demonstrate practical utility. In the comparative analysis of 76 nodes(i.e., with many singleton with SS=0), most conventional algorithms showed similar modularity (Q = 0.65) but near-zero silhouette scores, indicating weak cohesion despite acceptable partitioning. Components-based methods improved silhouette values to 0.60–0.62. In contrast, FLCA achieved the highest modularity (Q = 0.80) and mean silhouette score (0.78), producing clearer leader–follower clusters. The SSplot for the top 20 nodes across six clusters further showed positive silhouette values, supporting cluster stability. By integrating top-node filtering, FLCA-based clustering, algorithm comparison, network visualization, SSplots, Kano plots, and exportable reports, ClusterChoice provides a practical and reproducible tool for improving bibliometric cluster selection and network-quality assessment. • Compares 11 clustering algorithms for bibliometric network analysis. • Uses top-node filtering to reduce dense co-word network complexity. • Reports modularity, silhouette, ARI, and NMI with exportable visual outputs.