Edge-Conditioned Gat for Journal Ranking in Citation Networks

Traditional journal ranking metrics like the Journal Impact Factor (JIF) depend on raw citation counts, often overlooking citation context, interdisciplinary influence, and temporal dynamics, which restricts their ability to accurately evaluate scholarly impact. This paper introduces an edgeconditioned Graph Attention Networks (GAT) to refine JIF-based assessment by incorporating citation types weighted through adjustable schemes tailored to evaluation priorities. Trained on 11,054 journals from the Journal Citation Reports database, the model achieves strong predictive accuracy for JIF and produces rankings that closely align with JIF, SCImago Journal Rank, and Article Influence metrics. It uses an enhanced GAT variant by leveraging citation and publication counts, refining journals in rankings, reflecting their broader influence. In contrast to JIF's rigid, static methodology, this approach provides a configurable framework that adapts to diverse scholarly assessment needs, offering improvement over conventional citation-based metrics. By predicting JIF with high precision, this work bridges the gap between traditional bibliometric indicators and modern networkbased methods. It advances journal evaluation by combining predictive power with a deeper understanding of citation counts and their types with publication volume in one framework.