Dynamic Identification and Optimization of Key Opinion Leaders in Social Networks Using Advanced GCN Models

The rapid evolution of social media platforms has significantly increased the speed and reach of information dissemination, making the identification of Key Opinion Leaders (KOLs) crucial for shaping public opinion and maximizing influence. This paper presents a novel approach to identifying KOLs through the construction of dynamic social networks, employing advanced Graph Convolutional Networks (GCNs) and a pseudo-labeling mechanism. Pseudo-labeling is an innovative technique that automatically generates labels for nodes in a social network based on their characteristics, such as connectivity and influence level. By categorizing nodes into leader and edge nodes, we enhance the training data available for the dynamic GCN model, enabling it to better capture the underlying relationships within the network. By transforming user interaction data into a weighted social network and applying a community detection algorithm, we effectively generate a pool of candidate opinion leaders. The integration of dynamic GCN models, optimized through a Mahalanobis distance-based similarity measure, further refines the selection process, ensuring that the most influential leaders are identified. Experimental results on real-world datasets demonstrate that our method, outperforms existing algorithms in influence spread and leader identification accuracy, achieving a 12.35% improvement in KOL identification and enhancing precision in influence propagation and community detection, especially with fewer KOLs or clusters.