A novel positive-dominated negative synthesizing (PDNS) strategy that mitigates the over-fitting issue caused by false negatives in hard negative sampling for recommender systems.
