Enhanced Bayesian Personalized Ranking for Robust Hard Negative Sampling in Recommender Systems
The proposed Hard-BPR loss function mitigates the influence of false negatives in hard negative sampling, improving the robustness and effectiveness of recommendation model training.