High-quality community detection in complex networks based on node influence analysis

Numerous real-world systems can be represented as complex networks of some sort. Mining communities can aid in the analysis of the structural traits and functionalities of complex networks, as community structure is a crucial component of these networks. Seed nodes are crucial as they can influence the outcome of the community detection algorithm and help identify community structures. So far, several techniques have been presented to identify communities based on seed nodes as influential nodes. Nevertheless, most of these studies develop techniques in which seed nodes are considered as community centers. Also, these works include shortcomings such as instability, weak scalability, and low accuracy, which lead to the identification of different communities in successive executions. With this motivation, this paper develops a strategy based on Node Influence analysis for high-quality Community Detection in complex networks (NICD). NICD provides a new centrality metric to calculate node importance, which is based on the probability of information transfer between each pair of nodes and applying the influence scope. Unlike conventional weighting techniques, NICD compares the node influence vector using the Pareto archive to select a set of non-dominant nodes as influential nodes. According to the proposed centrality metric, the influential nodes are used to determine the centers of the communities, and then the final community structures are created using the k-medoid algorithm. Lastly, numerical simulations are applied on several real-world networks to verify the effectiveness of the proposed influential node-based community detection strategy. The results show the success of the proposed strategy by comparing the findings from different aspects.