Learning to Collaborate via Structures: Cluster-Guided Item Alignment for Federated Recommendation

TL;DR

Cluster-Guided FedRec framework is proposed, a framework that transforms uploaded embeddings into compact cluster labels and achieves the effective injection of global collaborative signals into local item representations without transmitting full embeddings.

Abstract

Federated recommendation facilitates collaborative model training across distributed clients while keeping sensitive user interaction data local. Conventional approaches typically rely on synchronizing high-dimensional item representations between the server and clients. This paradigm implicitly assumes that precise geometric alignment of embedding coordinates is necessary for collaboration across clients. We posit that establishing relative semantic relationships among items is more effective than enforcing shared representations. Specifically, global semantic relations serve as structural constraints for items. Within these constraints, the framework allows item representations to vary locally on each client, which flexibility enables the model to capture fine-grained user personalization while maintaining global consistency. To this end, we propose Cluster-Guided FedRec framework (CGFedRec), a framework that transforms uploaded embeddings into compact cluster labels. In this framework, the server functions as a global structure discoverer to learn item clusters and distributes only the resulting labels. This mechanism explicitly cuts off the downstream transmission of item embeddings, relieving clients from maintaining global shared item embeddings. Consequently, CGFedRec achieves the effective injection of global collaborative signals into local item representations without transmitting full embeddings. Extensive experiments demonstrate that our approach significantly improves communication efficiency while maintaining superior recommendation accuracy across multiple datasets.

Figures (9)

• Figure 1: Comparison of representative federated recommendation paradigms for communication. $O(\cdot)$ denotes the communication complexity.
• Figure 2: The framework of CGFedRec. Each client first trains a local recommendation model to learn item representations. The server aggregates the uploaded representations and discovers global item structures via clustering, producing cluster labels for all items. Instead of broadcasting global embeddings, the server only sends these lightweight cluster labels back to clients. Clients then use the labels as structural supervision to align item representations through contrastive learning, enabling global knowledge injection with significantly reduced communication cost.
• Figure 3: Heatmaps of item embeddings for 4 randomly selected clients in CGFedRec on the ML-100K dataset. Each heatmap visualizes 32 randomly selected items, with color intensity representing embedding values.
• Figure 4: Heatmaps of item-item similarity for 4 selected clients in CGFedRec on the ML-100K dataset. Each heatmap shows cosine similarity computed from item embeddings (Figure \ref{['fig:embeddings']}) for 32 randomly selected items, with color intensity representing similarity values.
• Figure 5: Effect of the temperature parameter on the performance of CGFedRec on the ML-100K dataset.