Knowledge graph QA often assumes a single system can access the whole graph, yet in practice organizations hold disjoint relation types while sharing entity identifiers, so no party sees a complete reasoning chain. FedV-KGQA evaluates multi‑hop QA on such vertically partitioned graphs. Each silo enriches its local graph and trains a KG embedding on its own triples. A server then concatenates the silo‑specific entity views, anchors the projected question at the topic entity, and ranks candidates by similarity. Raw triples and relation embeddings never leave their silos. Experiments reveal three findings: (1) federated fusion recovers most of the centralized accuracy, while a single silo performs poorly; (2) anchoring and enrichment matter more than the choice of embedding model; (3) the cheapest encoder depends on the target accuracy rather than parameter count. The poster contributes a cross‑experiment comparison, four design lessons, and an interactive prototype that runs real inference and traces the full pipeline per question on released checkpoints.
Review