Efficient Community Structure Identification in Social Networks Using K-Core Based Detection and Influential Information Diffusion Modeling

Abstract

This research introduces an innovative community discovery technique employing a weighted K-core methodology specifically designed for unguided graphs inside social media platforms.  The technique allocates weights for edges based upon users’ interactions—likes, responses, and shares—reflecting the intensity and character of social relationships.  Nodes are assessed for their coreness, indicating their importance and centrality inside the system.  A multi-phase method is suggested, commencing with data preparation and graph creation utilizing the social network fan page behaviour.  Edge recognition and weight computation techniques assess the significance of user interactions, classifying them as the original, insignificant, or core interest.  The K-core decomposition technique reveals densely linked subsections, signifying groups of power.  This methodology facilitates the categorization of users into highly or weakly linked clusters according to their local coreness, hence aiding in the differentiation between influencing and non-influential individuals.  The method is evaluated with 7 real-world Facebook datasets and contrasted against existing techniques including Info map, Louvain, and Label Spreading.  Experimental findings indicate that the suggested K-core technique attains enhanced accuracy, modularity, and F1-score, while ensuring scaling for vast networks.  The results validate its efficacy in identifying significant groups and user roles within evolving social frameworks. The proposed K-Core technique attains the best overall accuracy at 96.5%, surpassing all previous models.  Louvain attains 85.2%, Infomap registers 81.7%, Label Propagation secures 80.6%, Fast Greedy accomplishes 79.7%, Walktrap obtains 78.0%, while Leading Eigenvector encounters the lowest possible accuracy at 76.0%.