Polaris-JZ / PDRO

Popularity-aware Distributionally Robust Optimization for Recommendation System

This is the pytorch implementation of our paper

Popularity-aware Distributionally Robust Optimization for Recommendation System

Environment

• Anaconda 3
• Python 3.8.12
• Pytorch 1.7.0
• Numpy 1.21.2

The Micro-video Dataset

The Micro-video dataset, sourced from the Huawei micro-video App integrated into Huawei mobile phones, comprises a collection of user-item interactions spanning a month. This dataset encompasses a wide range of interactions with diverse micro-videos.

#user	#item	#Interaction
25,871	44,503	210,550

Training

Run the PDRO on Micro-video dataset:

sh run.sh micro_video lgn 1e-3 0.001 0.3 0.17 3 8 0.3 1 4 0.2 0 log_0 0

Inference (Including Group Evluation)

Infer the PDRO on Micro-video dataset:

sh inference.sh micro_video lgn 1e-3 0.001 0.3 0.17 5 8 0.3 1 4 0.2 0 log_0 1

This implementation is based on LightGCN.