Awesome

shadow-free-MIAs

This respository contains the source code of the IJCAI-24 paper "Shadow-Free Membership Inference Attacks: Recommender Systems Are More Vulnerable Than You Thought". Authors proposed a novel membership inference attack against recommender systems without shadow training.

Requirement

• torch == 2.1.1
• python == 3.9.18
• numpy == 1.26.0
• pandas == 2.1.3

Dataset

The experiments are evaluated on three benchmark datasets, i.e., MovieLens-1M, Amazon Beauty, and Ta-feng.

• For MovieLens-1M, download dataset from here, then put it into the path "/dataprocess/"
• For Amazon Beauty, download dataset from here, then put it into the path "/dataprocess/"
• For Ta-feng, download dataset from here, then put it into the path "/dataprocess/"

Recommender System

• Traditional recommender system
  • For Item-based Collaborative Filtering(ICF), follow this link to implement ICF.
• Advanced deep learning based recommender systems
  • NCF, follow this link to implement NCF.
  • BERT4Rec, follow this link to implement BERT4Rec.
  • Caser, follow this link to implement Caser.
  • GRU4Rec, follow this link to implement GRU4Rec.

Get started

The following command can be used to train shadow-free MIAs for both traditional recommender systems and advanced deep learning based recommender systems:

cd attack/SFMD/attackModel/
python beauty_Bert4Rec.py