This is the official code for [HGIB (Hierarchical Graph Information Bottleneck for Multi-Behavior Recommendation)].

📝 RecSys 2025

In this project, we propose a novel model-agnostic Hierarchical Graph Information Bottleneck (HGIB) framework for multi-behavior recommendation to effectively address these challenges. Following information bottleneck principles, our framework optimizes the learning of compact yet sufficient representations that preserve essential information for target behavior prediction while eliminating task-irrelevant redundancies. To further mitigate interaction noise, we introduce a Graph Refinement Encoder (GRE) that dynamically prunes redundant edges through learnable edge dropout mechanisms. We conduct comprehensive experiments on three real-world public datasets, which demonstrate the superior effectiveness of our framework.

The main prerequisites are listed below:

Python 3.9
torch
tensorboard

And the entire dependencies can be set up by run:

pip install -r requirements.txt

We provide the Taobao, Tmall and Jdata datasets in './data' folder.

Tmall and Jdata datasets are gathered from CRGCN and Taobao dataset is gathered from MBCGCN.

You can run the following script for preprocessing:

python ./data/preprocess.py

python ./src/main.py --dataset taobao --lr 5e-4 

python ./src/main.py --dataset tmall  --lr 5e-4 

python ./src/main.py --dataset jdata  --lr 5e-4 --alpha 0.5

Our code is developed based on MuLe.

@inproceedings{10.1145/3705328.3748073,
author = {Zhang, Hengyu and Shen, Chunxu and Sun, Xiangguo and Tan, Jie and Tan, Yanchao and Rong, Yu and Cheng, Hong and Yi, Lingling},
title = {Hierarchical Graph Information Bottleneck for Multi-Behavior Recommendation},
year = {2025},
isbn = {9798400713644},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3705328.3748073},
doi = {10.1145/3705328.3748073},
booktitle = {Proceedings of the Nineteenth ACM Conference on Recommender Systems},
pages = {155–164},
numpages = {10},
location = {Prague, Czech Republic},
series = {RecSys '25}
}