IFL: Mitigating Spurious Correlations for Self-supervised Recommendation

This is the pytorch implementation of our paper

Mitigating Spurious Correlations for Self-supervised Recommendation

Xinyu Lin, Yiyan Xu, Wenjie Wang, Yang Zhang, Fuli Feng

Environment

• Anaconda 3
• python 3.7.15
• pytorch 1.7.0
• numpy 1.21.5

Usage

Data

The experimental data of XING are in './data' folder.

Training

python main.py --model=$1 --dataset=$2 --lr=$3 --hidden_factor=$4 --batch_size=$5 --layers_u=$6 --layers_i=$7 --temp=$8 --alpha=$9 --beta=$10 --env=$11 --regs=$12 --regs_mask=$13 --dropout=$14 --batch_norm=$15 --log_name=$16 --gpu=$17

or use run.sh

sh run.sh model_name dataset lr hidden_factor batch_size DNN_layers_user DNN_layers_item temperature alpha beta n_env reg reg_mask dropout batchNorm log_name gpu_id

• The log file will be in the './code/log/' folder.
• The explanation of hyper-parameters can be found in './code/main.py'.
• The default hyper-parameter settings are detailed in './code/hyper-parameters.txt'.

Example

1. Train IFL on XING:

cd ./code
sh run.sh IFL XING 0.01 64 1024 [] [] 0.7 0.7 0.001 2 0.0001 0.0001 [0,0] 0 log 0