A Comparative Study of Group Profiling Techniques in Co-Authorship Networks

Abstract

Group profiling methods aim to construct a descriptive profile for communities in complex networks. The application of such methods in the analysis of co-authorship networks enables us to move forward in understanding the scientific communities, leading to new approaches to strengthen and expand scientific collaboration networks. This task is similar to the document cluster labeling task, which encourages the adaptation of cluster labeling methods for group profiling problems. In this work, we present a comparative study of group profiling and cluster labeling algorithms in a co-authorship network. A qualitative survey was conducted to evaluate the generated profiles, as well as the pros and cons of different profiling strategies, were analyzed with concrete examples. The results demonstrated a similar performance of both group profiling and cluster labeling methods.