A new hierarchical clustering algorithm to identify non-overlapping like-minded communities

Abstract: A network has a non-overlapping community structure if the nodes of the network can be partitioned into disjoint sets such that each node in a set is densely connected to other nodes inside the set and sparsely connected to the nodes out- side it. There are many metrics to validate the efficacy of such a structure, such as clustering coefficient, betweenness, centrality, modularity and like-mindedness. Many methods have been proposed to optimize some of these metrics, but none of these works well on the recently introduced metric like-mindedness. To solve this problem, we propose a be- havioral property based algorithm to identify communities that optimize the like-mindedness metric and compare its performance on this metric with other behavioral data based methodologies as well as community detection methods that rely only on structural data. We execute these algorithms on real-life datasets of Filmtipset and Twitter and show that our algorithm performs better than the existing algorithms with respect to the like-mindedness metric.