GP-FedRS: Improving Ranking Quality in Heterogeneous Federated Recommendation with Trainable Global Prototypes

Written by

in

GP-FedRS improves the robustness and ranking quality of heterogeneous federated recommendation by learning trainable global prototypes on the server and using dynamic-margin contrastive learning. The framework also employs difference-aware client selection to improve semantic coverage. Across four public benchmark datasets and eight strong baselines, it achieves the best HR and NDCG performance, with relative improvements of 4.82% to 11.36% over the strongest baseline.

Source:BIOENGINEER.ORG · Read Original · 前沿热点

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *