Enhanced Temporal Graph Neural Network for Predicting Future Citations on Academic Graphs: A Dual Clustering-Driven and Centrality-Guided Approach

RAIDS Lab Authors

Zhengyi Yang

Director

Details

Research Area

Tags

Resources

Abstract

As the volume of scientific publications grows, predicting the future citation counts of research papers is crucial for identifying influential studies and promising research directions. Existing graph neural network methods often rely solely on neighbor aggregation, which neglects the global heterogeneous nature of academic networks, thereby limiting their ability to fully capture the intricate relationships between different types of nodes and edges. This paper proposes a novel model, named Centrality-guided and Dual Clustering driven Heterogeneous Graph Network (CDCHGN) that addresses these limitations. CDCHGN captures both global and local structural dynamics of a dynamic academic graph using node centrality information, dual clustering techniques, and edge-type encoding to enrich semantic representation and effectively model heterogeneous relationships among various node types. Additionally, it incorporates time-injected attention to capture the temporal evolution of node features. Extensive experiments on real-world datasets demonstrate that our model achieves superior performance. Notably, on the APS dataset, CDCHGN achieves a 5.88% improvement in Mean Absolute Log Error (MALE) and a 6.87% improvement in Root Mean Squared Log Error (RMSLE) compared to the second-best models.

Author Affiliations

@article{zhang2025enhanced,
  title = {Enhanced temporal graph neural network for predicting future citations on academic graphs: A dual clustering-driven and centrality-guided approach},
  author = {Zhang, Tianming and Fang, Junkai and Chen, Xuanyu and Yang, Zhengyi and Cao, Bin},
  volume = {329},
  issn = {0950-7051},
  url = {http://dx.doi.org/10.1016/j.knosys.2025.114282},
  doi = {10.1016/j.knosys.2025.114282},
  journal = {Knowledge-Based Systems},
  publisher = {Elsevier BV},
  year = {2025},
  month = Nov,
  pages = {114282}
}