HDPMF

This is our implementation for dencentralized matrix factorization with heterogeneous differential privacy.

How to run

1. Download datasets

• ML-100K/ML-1M
• Douban
• Yelp

2. Preprocess datasets and put them in the folder /Data

• For default setting or changing hyperparmeters experiments

python dataprocess_for_default.py 

• For changing dataset sparsity experiments

python dataprocess_for_sparsity.py 

3. Run model

• Run Our model HDPMF :

python mf_hdp.py --data Data/ml-1m  --lr 0.01 --embedding_dim 10 --regularization 0.01  --stddev 0.1

• Run baseline method PDPMF:

python mf_sampling.py --data Data/ml-1m  --lr 0.01 --embedding_dim 10 --regularization 0.01  --stddev 0.1

• Run original MF without any noise:

python mf_nonprivate.py --data Data/ml-1m  --lr 0.01 --embedding_dim 10 --regularization 0.01 --stddev 0.1

Requirements

python: 3.8
sklearn: 1.0.2

Code Reference

• Neural Collaborative Filtering vs. Matrix Factorization Revisited