Graph-based modeling using association rule mining to detect influential users in social networks

•A proposed graph-based approach using association rule mining to detect influential users in social networks.•It includes a new diffusion-graph-based algorithm for influence maximization problem.•A new centrality measure, called completeness centrality, of identifying top nodes using all possible paths in the network.•Experimental results show the effectiveness and efficiency of our proposed measure and graph-based approach. Information diffusion is an important and attractive field of research in the area of social network analysis, and is at the heart of many studies and applications of knowledge extraction and prediction. Most of these studies have focused on identifying the most influential users and predicting user participation. Nevertheless, despite the extensive research efforts that have been made to tackle these issues, there is still a need for approaches based on association rules mining and graph theory. In this study, we contribute to research in this field by introducing a novel graph-based approach that applies association rules mining to detect influential users. We argue that users influence each other, and that it is possible to predict a user’s interests and participation based on previous interactions in the social network. We introduce new concepts and algorithms for more efficient characterization of influential users, and develop an effective approach for the discovery of influencers by using association rule techniques to extract the hidden relationships between users. To evaluate the feasibility and effectiveness of our approach, we propose a new centrality measure called the completeness centrality, and perform an evaluation based on a case study selected from the literature. We then evaluate the effectiveness of the proposed centrality measure by using the susceptible-infected-recovered model and the overlapping similarity measure. The results demonstrate that our measure is feasible and effective for use in identifying influential spreaders, based on a comparison with existing centrality measures such as degree, betweenness, closeness, and eigenvector methods. Finally, to illustrate the efficiency of our approach, experiments were run on 25 generated diffusion graphs, and the results showed that our approach could achieve a high level of performance in terms of computational time for large-scale networks.