A sentiment-aware one-class collaborative filtering method is proposed.
The method integrates user sentiment with ratings via fixed or learned mappings.
The method compares favorably with other models on three large multimedia datasets.
Improvements are consistent and independent from the negative class assumption.