FedCGR: Federated Cross-Domain Recommendation via Semantic ID Sequences
August 12, 2026
FedCGR enables cross-domain recommendation in federated settings by representing items as discrete semantic ID (SID) sequences derived from public metadata. This approach uses a shared vocabulary to align item spaces across clients, bypassing the need to exchange private user interactions or align domain-specific embeddings.
HOW THIS AFFECTS YOU
●
builderYou can implement privacy-preserving recommendation systems without sharing user behavioral data.
●
researcherYou can explore semantic-only bottlenecks in federated learning environments.