UC-FedRec

Source Code for WSDM2024 paper "User Consented Federated Recommender System Against Personalized Attribute Inference Attack"

Introduction

We propose user-consented federated recommender systems (UC-FedRec) against attribute inference attacks to meet the personalized privacy demands of clients. The framework learns a set of attribute information filters to eliminate sensitive information to protect clients' personal attributes from attackers' malicious inferences.

Reproduction

Package Dependencies

• numpy
• pandas
• scipy
• scikit-learn == 1.0.2
• torch == 1.7.1
• dgl == 0.6.1

Data Preparation

Please download the dataset used at Onedrive.

Unzip the file and put it under the root directory of this project.

Train UC-FedRec

UC-FedRec sample usage at Movielens dataset:

python main_UC_ML.py --layer_size [128] --batch_size 256 --embed_size 128 --Ks [10] --gpu 3 --lr 0.0001 --model_name sgd_model_run4_1.pkl

Citations

The details of this pipeline are described in the following paper. If you use this code in your work, please kindly cite it.

Miscellaneous

Please send any questions about the code and/or the algorithm to qhuaf@connect.ust.hk.