This is the source code for the LLPAUC. The code references Recbole (https://github.com/RUCAIBox/RecBole)
The candidate loss functions are: CCL, TP_Point_TP, TP_Point_OP, BPR, BCE, softmax
CCL:Cosin Constractive Loss
TP_Point_TP:LLPAUC Loss in our paper
TP_Point_OP:OPAUC Loss in our paper
BPR:Bayesian Personalized Ranking Loss
BCE:Binary Cross-Entropy Loss
softmax:Softmax Cross-Entropy Loss(SCE in our paper)
The candidate datasets are: adressa_clean,adressa_noise,yelp_clean,yelp_noise,amazon_book_clean,amazon_book_noise
In order to reproduce the results reported in our paper, we set the default hyper-parameters for our paper. For example, the command to obtain the LLPAUC results for amazon_book_clean dataset is
python -u run_main.py --dataset=amazon_book_clean --loss=TP_Point_TP