EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Subgraph-Causal Invariance for Robust Link Prediction under Structural Distribution Shifts

Emma Johansen

International Journal of Business and Social Science Studies · 10.09.2026 · 6 (2) · 142-151

DOI: 10.66795/ijbsss67 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Business and Social Science Studies, online erschienen am 10.09.2026. Verfasst von Emma Johansen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Link prediction is a fundamental task in graph machine learning with extensive applications in social network analysis, bioinformatics, and recommendation systems. While Graph Neural Networks have achieved remarkable success in this domain under the assumption of independent and identically distributed data, their performance severely degrades when deployed in real-world scenarios characterized by structural distribution shifts. These shifts, which manifest as variations in node degree distributions, localized network densities, or community structural topologies between training and testing environments, confound traditional predictive models that rely on superficial topological correlations rather than true causal relationships. To address this critical vulnerability, this paper introduces a comprehensive theoretical and methodological framework centered on subgraph-causal invariance. By modeling the link existence problem through the lens of structural causal models, we distinguish between invariant causal subgraphs, which dictate the true mechanisms of edge formation, and variant environmental subgraphs, which encapsulate spurious correlations specific to the training distribution. The proposed framework extracts localized subgraph topologies around target node pairs and applies an invariant risk minimization objective across synthetically generated or naturally occurring graph environments. Through rigorous theoretical analysis and extensive empirical validation across multiple domain datasets, we demonstrate that isolating causal subgraph structures yields unprecedented robustness against severe out-of-distribution structural shifts. The methodology significantly outperforms state-of-the-art heuristic and neural link prediction algorithms, ensuring stable, reliable, and interpretable edge inferences even when testing distributions deviate drastically from training conditions.

Zitieren

APA

Johansen, E. (2026). Subgraph-Causal Invariance for Robust Link Prediction under Structural Distribution Shifts. International Journal of Business and Social Science Studies, 6(2), 142-151. https://doi.org/10.66795/ijbsss67

Vancouver

Johansen E. Subgraph-Causal Invariance for Robust Link Prediction under Structural Distribution Shifts. International Journal of Business and Social Science Studies. 2026;6(2):142-151. doi:10.66795/ijbsss67

BibTeX
@article{Johansen2026,
  author = {Johansen, Emma},
  title = {Subgraph-Causal Invariance for Robust Link Prediction under Structural Distribution Shifts},
  journal = {International Journal of Business and Social Science Studies},
  year = {2026},
  volume = {6},
  number = {2},
  pages = {142--151},
  doi = {10.66795/ijbsss67}
}
Kostenlos starten →