Hybrid Graph Neural Networks and Explainable AI for Predictive Analytics in Dynamic Knowledge Graphs: A Comprehensive Review

Devendra Rawat, Garima Talwar, Hemant Lamba

Abstract


Dynamic Knowledge Graphs (DKGs) have emerged as powerful abstractions for representing complex, evolving topological structures and time-varying semantic relationships across diverse domains, including financial fraud detection, biomedical entity interaction, dynamic recommender systems, and cybersecurity threat intelligence. While Graph Neural Networks (GNNs) have achieved state-of-the-art predictive performance on relational learning tasks, standard architectures struggle with the dual challenges of capturing temporal fine-grained dynamics and providing interpretable, transparent reasoning mechanisms. This comprehensive review paper synthesizes the paradigm shift toward Hybrid Graph Neural Networks combined with Explainable Artificial Intelligence (XAI) tailored specifically for predictive analytics in dynamic knowledge graphs. We systematically classify hybrid architectural taxonomies—focusing on spatial-temporal GNNs, recurrent graph convolutional networks, temporal graph attention mechanisms, and continuous-time neural ordinary differential equations (Neural ODEs). Furthermore, we evaluate dynamic attribution methods, post-hoc subgraph explainers, and self-explaining temporal architectures. Through extensive synthesis of experimental metrics, benchmark evaluation frameworks, and cross-domain applications, we analyze the trade-offs between predictive fidelity and explanation sparsity. Finally, we highlight persistent open challenges, including scalability limits, temporal out-of-distribution generalization, and dynamic concept drift, outlining concrete future directions for trustworthy dynamic graph intelligence.

KEYWORDS: Dynamic Knowledge Graphs, Hybrid Graph Neural Networks, Explainable AI (XAI), Predictive Analytics, Temporal Graph Attention, Subgraph Attribution

Full Text:

PDF 113-128

Refbacks

  • There are currently no refbacks.