Clustering Scientometrics of Computer Science Journals for Subarea Decomposition

Scientometrics indicators vary widely across subareas of the Computer Science (CS) discipline. Most researchers have previously analyzed scientometrics data specific to a particular subfield or a few subfields. More popular subareas lead to high scientometrics, and others have lower values. This work considers seven diversified CS subareas and six commonly used scientometrics indicators. First, we study the varying range of chosen scientometrics indicators of various subareas of the CS discipline. We explore the correlation patterns of these six indicators. Then, we consider a few combinations of these indicators and apply K-means clustering to decompose the pattern space. Correlation findings indicate that though the highly correlated indicators vary for most subfields, no single indicator can be considered equally suitable for all the subareas. The K-means clustering results show distinctive patterns across subfields, which are stable across K. The clustered subfield-specific indicators are quite distinct across subfields. This knowledge can be used as a signature for partitioning the subarea-specific indicators.

Keywords: Scientometrics, Bibliometrics, Publications, K-means, Clustering, Computer Science, Subarea Indicators, Machine Learning.