The rapid growth of social networks has introduced complex challenges in detecting and analyzing their underlying community structures. Traditional community detection methods often fall short in capturing the dynamic, overlapping, and hierarchical nature of these networks. This study proposes a hybrid framework that integrates Graph Convolutional Networks (GCNs), Recurrent Neural Networks (RNNs), and attention mechanisms within a semi-supervised learning paradigm to enhance community detection in social networks. By leveraging both labeled and unlabeled data, the proposed GCN-RNN-Attention model effectively captures both structural and temporal dynamics of evolving networks. Experimental evaluations on benchmark datasets—including Karate Club, Cora, CiteSeer, Facebook, and PubMed—demonstrate the superiority of the proposed model in terms of modularity, accuracy, normalized mutual information (NMI), and F1-score compared to traditional and state-of-the-art methods. Ablation studies further confirm the contribution of each component to overall performance, and scalability analysis proves the method’s robustness across datasets of varying sizes. This research contributes to advancing community structure analysis with implications for applications in marketing, fraud detection, epidemic modeling, and social influence analysis.
