APR enhances the pairwise ranking method BPR by performing adversarial training. To illustrate how it works, APR on FISM is implemented here by adding adversarial perturbations on the matrices P and Q.
Primary LanguagePython