newlei / FDA

Improving Recommendation Fairness via Data Augmentation-WWW23

FDA

WWW 2023 Conference

Paper title: Improving Recommendation Fairness via Data Augmentation arxiv WWW

Prerequisites

• PyTorch 1.7
• Python 3.5
• NVIDIA GPU + CUDA CuDNN

Getting Started

Installation

• Clone this repo

Train/test

• Train FDA_BPR on MovieLens:

cd FDA_anonymous/fda_bpr_ml
python main.py

• Train FDA_GCCF on MovieLens:

cd FDA_anonymous/fda_gccf_ml
python main.py

• Train FDA_BPR on LastFM:

cd FDA_anonymous/fda_bpr_lastfm
python main.py

• Train FDA_GCCF on LastFM:

cd FDA_anonymous/fda_gccf_lastfm
python main.py

Note: The results of FDA will be output on the terminal after the training.

ERROR:

1. The error " 'weight' must be 2-D" occurred due to inconsistent versions of the Pytorch version.

• Solution:

gender = F.embedding(u_batch,self.users_features)
male_gender = gender.type(torch.BoolTensor)
female_gender = (1-gender).type(torch.BoolTensor)

Replace the above code with the following code:

gender = F.embedding(u_batch,torch.unsqueeze(self.users_features,1)).reshape(-1)
male_gender = gender.type(torch.BoolTensor).cuda()
female_gender = (1-gender).type(torch.BoolTensor).cuda()