Determining Leaders and Communities on Networks Using Neighborhood Similarity

Abstract

Networks are used to represent interactions in a wide variety of fields, like biology, sociology, chemistry, and more. They have a great deal of salient information contained in their structures, which have a variety of applications. One of the important topics of network analysis is finding influential nodes. These nodes are of two kinds -leader nodes and bridge nodes. In this study, we propose an algorithm to find strong leaders in a network based on a revision of neighborhood similarity. This leadership detection is combined with a neighborhood intersection clustering algorithm to produce high quality communities for various networks. We also delve into the structure of a new network, the Houghton College Twitter network, and examine the discovered leaders and their respective followers in more depth than which is frequently attempted for a network of its size. The results of the observations on this and other networks demonstrate that the community partitions found by this algorithm are very similar to those of ground truth communities.

The Effects of Centrality Ordering in Label Propagation for Community Detection

“…Betweenness centrality on the other hand evaluates nodes based on the number of shortest paths passing through them [8]. Leadership quality introduced by B. Valyou et al in [9] weights nodes based on their aggregate neighborhood similarity with each node in their neighborhood. Alpha centrality is a modification of the eigenvector centrality function which weights nodes based on their degree and the degree of their neighbors [10].…”