Discovering centrality clusters in social and interaction networks using AI-driven association analysis

Abstract The social network is an interconnected online web of people spanning across the digital space for sharing information, reflecting opinions, and building societal trends. This necessitates quantification of relative prominence of an individual in the online community. The significance of a node is measured by centrality metrics, with different indicators accentuating different characteristics of the network. Though there are a myriad of centrality indicators however, it would be interesting to explore which centrality indicators are highly correlated. The literature survey shows that extensive study is still required to establish the network-specific choice of centrality measures for defining node importance. In our work, we have analysed different classes of networks by calculating various centrality indicators. We have computed the pair-wise correlation of the centrality indicators for each network. This is followed by setting threshold value to find the strongly associated centrality indictors. These indicators are further partitioned into clusters, which are validated by comparing the top 50 ranked nodes based on the strongly correlated centrality measures. We have proposed technique to find the similarity in clustering of centrality indictors for each category of networks. Also, we have explored the similarity of centrality indices irrespective of the network class for generalization. Our findings have established that each member of a cluster of correlated centralities is capable of serving as a suitable indicator for detecting the influential nodes based on the structural properties of a specific class of network. Furthermore, we employed association rule mining, a machine learning approach, to uncover frequently co-occurring centrality indicators that exhibit consistent patterns across networks. This AI-driven analysis provides deeper insight into the interdependencies among centrality measures, facilitating more efficient and interpretable network analysis.