Community Detection and Characteristics Analysis of Academic Network

Clustering and categorization of academic network, especially collaboration network, is always an interesting question, in that it can provide many insight into academic fields. While traditional methods, such as spectral clustering, are based on the assumption that each node can be assigned to only one clustering, more and more observations suggest that communities in academic network probably overlap with each other, or even nested under each other. This motivates us to apply a novel overlapping community detection algorithm on a large collaboration networks we obtain from arXiv with ground-truths. In this paper, we will examine the community detection on our collaboration network for deep insight into the internal structure that was hard to see in the past. By studying leadership characteristics of collaboration network and how leaders distribute among overlapping communities in various years, we not only get more insight into the structure of collaboration network, but find some general disagreement between detected communities and ground-truth as well, which may guide more future model building work.