Preference Diffusion for Recommdation (ICLR 2025)

Shuo Liu^1,2, An Zhang^2*, Guoqing Hu³, Hong Qian¹, Tat-Seng Chua²,
¹East China Normal University, ²National University of Singapore,
³University of Science and Technology of China, (*Correspondence )

---

😸 Welcome to PreferDiff, this is a implementation of Preference Diffusion for Recommendation

1️⃣ ​ Guide for Running PreferDiff

🚶‍♂️ Single GPU

python main.py --model=PreferDiff --sd=O --td=O --loss_type=cosine  --lamda=0.4 --w=2 --hidden_size=3072  --ab=iids

🏃 Multi-GPU

CUDA_VISIBLE_DEVICES=0,1,2,3 accelerate launch --main_process_port=12330 main.py --model=PreferDiff --sd=O --td=O --loss_type=cosine  --lamda=0.4 --w=2 --hidden_size=3072 --ab=iids

2️⃣ Best Hyperparameters

Dataset	learning rate	Weight Decay	lambda	w	Embedding Size
Sports	1e-4	0	0.4	2	3072
Beauty	1e-4	0	0.4	8	3072
Toys	1e-4	0	0.6	6	3072

3️⃣ Guide for Running Baselines

CUDA_VISIBLE_DEVICES=0,1,2,3 accelerate launch --main_process_port=12330 main.py --model=SASRec --sd=O --td=O 

Bibtex

@inproceedings{Liu2024PreferDiff,
author = {Shuo Liu, An Zhang, Guoqing Hu, Hong Qian, Tat-seng Chua},
booktitle = {Proceedings of the 13th International Conference on Learning Representations},
title = {Preference Diffusion for Recommendation},
year = {2024},
address={Singapore}
}