Search papers, labs, and topics across Lattice.
This paper introduces FedCGR, a federated generative cross-domain recommendation framework that addresses the challenges of aligning item spaces in privacy-sensitive environments by leveraging a stable semantic item language. By representing items as discrete semantic ID sequences and incorporating local collaborative filtering evidence through a reliability-aware semantic interface, FedCGR effectively overcomes the limitations of traditional federated averaging and domain heterogeneity. Experimental results across six Amazon cross-domain scenarios demonstrate that FedCGR outperforms existing federated generative baselines and achieves competitive results against leading sequential and federated CDR methods.
FedCGR revolutionizes cross-domain recommendations by enabling effective item alignment without compromising user privacy or requiring extensive data sharing.
Cross-domain recommendation (CDR) transfers preference knowledge across related domains, but federated deployment makes cross-domain alignment difficult because the behavioral anchors that align item spaces, such as overlapping users and shared interaction signals, are often sparse, unavailable, or privacy-sensitive across clients. To address this tension, we revisit federated CDR as generation over a stable semantic item language. By representing items as discrete semantic ID (SID) sequences derived from public item-side metadata, cross-domain item alignment is induced by a shared vocabulary rather than by exchanging private interactions or aligning domain-specific embeddings. Directly federating SID-based generators, however, introduces two design constraints: the SID tokenizer must remain fixed to preserve cross-client token consistency, which creates a semantic-only bottleneck because local collaborative filtering (CF) signals cannot be globally shared or aligned; meanwhile, standard federated averaging can cause negative transfer under domain heterogeneity. To overcome these constraints, we propose FedCGR, a federated generative CDR framework that keeps the item language stable and makes adaptation explicit. FedCGR injects local CF evidence through a reliability-aware semantic interface and trains a prototype-personalized generator that selectively aggregates shared parameters according to domain relatedness while keeping domain-specific quantities local. Experiments on six Amazon cross-domain scenarios show that FedCGR consistently outperforms federated generative baselines and achieves competitive performance against strong sequential and federated CDR methods under both full-ranking and sampled evaluation protocols.
