跳转至

文章背景与核心概要

跨域推荐（CDR）旨在相关领域之间迁移偏好知识，但在联邦部署架构下，这一过程面临着巨大的阻碍。传统的跨域推荐严重依赖重叠用户和共享交互信号等“行为锚点”来对齐不同领域的物品空间，然而在联邦学习场景中，这些锚点往往由于隐私保护限制而变得稀疏、不可用或受到严格限制。

为了突破这些挑战，本文提出了 FedCGR 框架，将联邦跨域推荐重新构想为基于稳定语义物品语言的生成式任务。该方法将物品表示为从公开元数据中提取的离散语义 ID（SID）序列，从而利用共享词表自然地驱动物品对齐，而无需交换私有的用户交互数据。为了克服语义瓶颈以及由领域异质性带来的负迁移问题，FedCGR 引入了可靠性感知语义接口和原型个性化生成器。在六个亚马逊跨域场景上的实验表明，FedCGR 的表现超越了现有的联邦生成式基线，并能够与强大的序列推荐及联邦跨域推荐方法相媲美。

FedCGR: Federated Cross-Domain Generative Recommendation

Summary

Cross-domain recommendation (CDR) facilitates transferring preference knowledge across related domains, but doing so under a federated deployment presents significant hurdles. Behavioral anchors—such as overlapping users and shared interaction signals that traditionally align item spaces—are often sparse, unavailable, or restricted by privacy concerns across different clients.

Cross-domain recommendation (CDR) facilitates transferring preference knowledge across related domains, but doing so under a federated deployment presents significant hurdles. Behavioral anchors—such as overlapping users and shared interaction signals that traditionally align item spaces—are often sparse, unavailable, or restricted by privacy concerns across different clients.

To overcome these challenges, FedCGR re-envisions federated CDR as a generative task over a stable semantic item language. By treating items as discrete semantic ID (SID) sequences derived from public metadata, item alignment is naturally driven by a shared vocabulary rather than through the exchange of private user interactions. To combat the semantic bottleneck and negative transfer caused by domain heterogeneity, FedCGR introduces a reliability-aware semantic interface and a prototype-personalized generator. Experiments on six Amazon cross-domain scenarios demonstrate that FedCGR outperforms existing federated generative baselines and holds its own against strong sequential and federated CDR methods.

To overcome these challenges, FedCGR re-envisions federated CDR as a generative task over a stable semantic item language. By treating items as discrete semantic ID (SID) sequences derived from public metadata, item alignment is naturally driven by a shared vocabulary rather than through the exchange of private user interactions. To combat the semantic bottleneck and negative transfer caused by domain heterogeneity, FedCGR introduces a reliability-aware semantic interface and a prototype-personalized generator. Experiments on six Amazon cross-domain scenarios demonstrate that FedCGR outperforms existing federated generative baselines and holds its own against strong sequential and federated CDR methods.

Paper Metadata

  • arXiv ID: arXiv:2608.10929 [cs.AI]
  • Authors: Zhuodong Liu, Hugen Lv, Xiangyu Li, Bohan Guo, Peiyu Hu
  • Submitted: August 11, 2026
  • Accepted Venue: CIKM 2026 (10 pages, 5 figures, 6 tables)
  • License: Creative Commons Attribution 4.0 International license icon

Paper Metadata

  • arXiv ID: arXiv:2608.10929 [cs.AI]
  • Authors: Zhuodong Liu, Hugen Lv, Xiangyu Li, Bohan Guo, Peiyu Hu
  • Submitted: August 11, 2026
  • Accepted Venue: CIKM 2026 (10 pages, 5 figures, 6 tables)
  • License: Creative Commons Attribution 4.0 International license icon

Abstract

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.

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 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.

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.

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.