Power on graph: Mining power relationship via user interaction correlation

• Published in: Expert systems with applications 2025-04, Vol.270, p.126348, Article 126348
• Main Authors: Zang, Yilong, Ren, Lingfei, Wu, Junhang, Xiao, Yilin, Hu, Ruimin
• Format: Article
• Language: English
• Subjects: Graph neural network; Power relationship mining; Social network
• ISSN: 0957-4174
• DOI: 10.1016/j.eswa.2024.126348

Power relationships cannot be ignored in the social network, and mining them benefits a wide range of valuable applications, such as company management and leadership analysis. The focus of existing approaches is on how to explore social relationships more efficiently while ignoring the uniqueness of power relationships. In this paper, we first identify two unique challenges in power relationship mining: (1) User behavior. Communication between a leader and different subordinates tends to be highly variable. (2) Interaction structure. Power relationships are often interconnected like a pyramid shape but are overlapped in social interactions. Then we find and verify the existence of significant properties of power relationships’ correlation patterns on social networks to overcome the above challenges. Then a novel GNN model PRM-GNN to mine power relationships efficiently is proposed. We validate and illustrate the effectiveness and explainability of PRM-GNN using two real-world datasets. PRM-GNN achieves a 3.0% improvement in the F1-score compared to the State-of-the-art baseline on the Coauthor dataset and a 7.3% improvement on the Enron dataset. •A novel PRM-GNN model is proposed to mine power relationships in social networks.•PRM-GNN addresses complex user behavior and overlapping power structures.•User-Interaction-User (UIU) union representation improves power relationship mining.•Graph attention mechanism enhances model explainability and performance.•PRM-GNN outperforms state-of-the-art methods on real-world datasets.