Clustering Scientific Publications Based on Citation Relations: A Systematic Comparison of Different Methods

Area of Science:

• Bibliometrics
• Network Science
• Data Science

Background:

• Clustering methods are crucial for identifying research areas in bibliometric studies.
• These methods group publications based on citation network relationships.
• Numerous graph partitioning and community detection techniques exist in network science.

Purpose of the Study:

• To systematically compare the performance of various clustering methods for citation networks.
• To analyze the statistical properties and expert-based assessment of clustering results.
• To identify optimal methods for research field identification in bibliometrics.

Main Methods:

• Systematic comparison of numerous clustering algorithms on diverse citation networks (small to large scale).
• Statistical analysis of clustering results.
• Expert-based evaluation focusing on scientometrics publications.

Main Results:

• A trade-off exists between desirable properties for effective publication clustering.
• Map equation methods demonstrated superior performance in the comparative analysis.
• Findings suggest map equation methods warrant increased attention from bibliometric researchers.

Conclusions:

• Map equation methods offer a promising approach for research field identification.
• The study provides valuable insights into the strengths and weaknesses of different bibliometric clustering techniques.
• Further investigation into map equation methods is recommended for the bibliometric community.