# 對照組

### influence Injection
[influence function based data poisoning attacks to top-n recommender systems](https://arxiv.org/pdf/2002.08025.pdf)

1. 算出各item的「平均分數」，並找出平均分數最低的當作target item
2. 找出最受歡迎的的attribute
3. 將target item分別植入前n個最受歡迎的attribute

### max loss

[Adversarial Personalized Ranking for Recommendation](https://dl.acm.org/doi/pdf/10.1145/3209978.3209981?casa_token=NiS0mOREZAcAAAAA:FUXu-hDf4uzc77MFUPh7CyFSKGf-4t4mrrh2RBgRMTH0_rdfHBWLBgxveyiHQkqUK80ftVsj_NF36w)

1. 利用original data的資料找出每輪的loss
2. 找出loss最大的user-item pair
3. 插入到validation set