Abstract
A ranking service receives sparse request input and ranks candidates using a hybrid graph-neural and semantic-evidence architecture. The service computes graph-derived relevance using a Graph Neural Network (GNN), retrieves Resource Description Framework-star (RDF-star) statement-level semantic-relationship evidence, and computes a semantic score from the retrieved evidence. When direct semantic evidence is unavailable, bounded runtime fallback uses attenuated one-hop neighbor evidence. The service blends the graph-derived relevance and semantic score to produce ranked results with score decomposition and provenance indicators. The architecture improves semantic-context coverage and explainability for sparse enterprise graphs while preserving baseline neural-ranking behavior.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Mikhailov, Vladimir and Soundaramourty, Raja, "HYBRID GRAPH-NEURAL NET AND RESOURCE DESCRIPTION FRAMEWORK-STAR (RDF-STAR) SEMANTIC SIGNAL RANKING WITH SPARSE-SEMANTICS RECOVERY AND PROVENANCE-AWARE OUTPUT", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/11351