Federated Knowledge Graph Question Answering via Vertical Partitioning
September 15, 2026
FedV-KGQA enables multi-hop question answering across disjoint, vertically partitioned knowledge graphs without moving raw triples from local silos. The system achieves accuracy levels near centralized models by using entity anchoring and local embedding enrichment.
HOW THIS AFFECTS YOU
●
builderYou can build QA systems over private, distributed datasets without violating data sovereignty.
●
researcherThe findings highlight that anchoring and enrichment are more critical to federated accuracy than the specific embedding model used.