junkangwu / Adap_tau

[WWW 2023] Official code of "Adap-$\tau$: Adaptively Modulating Embedding Magnitude for Recommendation"

Adap-$\tau$: Adaptively Modulating Embedding Magnitude for Recommendation

This is the PyTorch implementation for our WWW 2023 paper.

Jiawei Chen, Junkang Wu, Jiancan Wu, Sheng Zhou, Xuezhi Cao, Xiangnan He. 2023. Adap-$\tau$: Adaptively Modulating Embedding Magnitude for Recommendation arxiv link

Dependencies

• pytorch==1.11.0
• numpy==1.21.5
• scipy==1.7.3
• torch-scatter==2.0.9

Training model:

• mkdir log
• cd bash

MF

yelp2018

# Adap_tau_0
bash Adap_tau_novel.sh  yelp2018 1e-4 1e-3 3 1024 2048 drop 1.0 1.0 uniform_gpu 0 100 cosine mf weight_v0
# Adap_tau
bash Adap_tau_novel.sh  yelp2018 1e-4 1e-3 3 1024 2048 drop 1.0 1.0 uniform_gpu 0 100 cosine mf weight_mean

amazon-book

# Adap_tau_0
bash Adap_tau_novel.sh amazon-book 1e-3 1e-7 3 1024 2048 nopdrop 1.0 1.0 uniform_gpu 0 100 cosine mf weight_v0
# Adap_tau
bash Adap_tau_novel.sh amazon-book 1e-3 1e-7 3 1024 2048 nopdrop 1.0 1.0 uniform_gpu 0 100 cosine mf weight_mean

gowalla

# Adap_tau_0
bash Adap_tau_novel.sh gowalla 1e-4 1e-9 3 1024 2048 drop 0.9 0.25 uniform_gpu 0 100 cosine mf weight_v0
# Adap_tau
bash Adap_tau_novel.sh gowalla 1e-4 1e-9 3 1024 2048 drop 0.9 0.25 uniform_gpu 0 100 cosine mf weight_ratio

LightGCN

yelp2018

# Adap_tau_0
bash Adap_tau_novel.sh  yelp2018 1e-3 1e-1 3 1024 2048 drop 1.0 1.0 no_sample 0 100 nocosine lgn weight_v0
# Adap_tau
bash Adap_tau_novel.sh  yelp2018 1e-3 1e-1 3 1024 2048 drop 1.0 1.5 no_sample 0 100 nocosine lgn weight_mean

amazon-book

# Adap_tau_0
bash Adap_tau_novel.sh amazon-book 1e-4 1e-1 3 1024 2048 nopdrop 1.0 1.0 no_sample 0 100 nocosine lgn weight_v0
# Adap_tau
bash Adap_tau_novel.sh amazon-book 1e-4 1e-1 3 1024 2048 nopdrop 1.0 1.0 no_sample 0 100 nocosine lgn weight_mean

gowalla

# Adap_tau_0
bash Adap_tau_novel.sh gowalla 1e-3 1e-5 3 1024 2048 nopdrop 0.8 0.6 no_sample 0 100 nocosine lgn weight_v0
# Adap_tau
bash Adap_tau_novel.sh gowalla 1e-3 1e-5 3 1024 2048 nopdrop 0.8 0.6 no_sample 0 100 nocosine lgn weight_mean

The training log is also provided. The results fluctuate slightly under different running environment.

For any clarification, comments, or suggestions please create an issue or contact me ([e-mail]).