UltraGCN

This is our Pytorch implementation for our CIKM 2021 paper:

Kelong Mao, Jieming Zhu, Xi Xiao, Biao Lu, Zhaowei Wang, Xiuqiang He. UltraGCN: An Ultra Simplification of Graph Convolutional Networks for Recommendation. Paper in arXiv.

Introduction

In this work, we propose an ultra-simplified formulation of GCN, dubbed UltraGCN. UltraGCN skips explicit message passing and directly approximate the limit of infinite message passing layers.

Environment Requirement

The required packages are as follows:

• python: 3.7.9
• pytorch 1.4.0
• numpy: 1.19.2
• scipy: 1.1.0
• tensorboard: 2.4.0

Code

• main.py: All python code to reproduce UltraGCN
• dataset_name_config.ini: The configuration file which includes parameter settings for reproduction on the specific dataset.

python main.py --config_file dataset_config.ini

Reproduction

See benchmarks folder to reproduce the results. For example, we show the detailed reproduce steps for the results of UltraGCN on the AmazoonBooks dataset in UltraGCN_amazonbooks_x0.md file.

Results