Discriminative-Invariant Representation Learning for Unbiased Recommendation

Introduction

In this project, we provide a novel discriminative-invariant representation learning (DIRL) method for unbiased recommendation.

Environment

We provide the environment that our code depends on in DIRL_env.ymal. To install the conda environment, run

conda env create -f DIRL_env.ymal

Dataset

1. Each line contains user ID, item ID, and label type (i.e., positve or negative).
2. We split Yahoo!R3 for training, validation, and testing (i.e., yahooR3t4p5_train for training, uni_yahooR3t4p5_val for validation, and uni_yahooR3t4p5_test for testing).

Run the Code

python main.py --data_name=yahooR3

Notes

Only the code for the performance comparison experiment is available to the public at this time. The other experiments will be shared once they have been organized and finalized.