Decentralized Multi-Target Cross-Domain Recommendation for Multi-Organization Collaborations

[arXiv] This is an implementation of Decentralized Multi-Target Cross-Domain Recommendation for Multi-Organization Collaborations

• (a) User-Aligned (b) Item-Aligned Decentralized Multi-Target Cross-Domain Recommendation (DMTCDR) for Multi-Organization Collaborations.

• The Learning and Prediction stages of Multi-Target Assisted Learning (MTAL).

Requirements

See requirements.txt

Instruction

• Global hyperparameters are configured in config.yml
• Experimental setup are listed in make.py
• Hyperparameters can be found at process_control() in utils.py
• organization.py define local initialization, learning, and inference of one organization
• assist.py demonstrate Multi-Target Assisted Learning (MTAL) algorithm
  • make_dataset() compute and distribute the pseudo-residual to all organizations
  • update() gather other domains' output and compute gradient assisted learning rate and gradient assistance weights
• The data are split at split_dataset() in data.py

Examples

• Train Joint (ML1M, User-Aligned, Explicit Feedback, MF, without side information)

  python train_recsys_joint.py --control_name ML1M_user_explicit_mf_0_genre_joint

• Test Alone (ML1M, Item-Aligned, Implicit Feedback, NCF, with side information)

  python test_recsys_alone.py --control_name ML1M_item_implicit_nmf_1_random-8_alone

• Train MDR (ML1M, User-Aligned, Explicit Feedback, MLP, without side information)

  python train_recsys_mdr.py --control_name ML1M_user_explicit_mlp_0_genre_mdr

• Train DMTMDR (Douban, User-Aligned, Explicit Feedback, AAE, without side information, with gradient assisted learning rate $\eta_k=0.3$, without gradient assistance weights)

  python train_recsys_assist.py --control_name Douban_user_explicit_ae_0_genre_assist_constant-0.3_constant

• Test DMTMDR (Amazon, User-Aligned, Implicit Feedback, AAE, without side information, with gradient assisted learning rate $eta_k=0.1$, with gradient assistance weights, with partial alignment (0.5), with privacy enhancement (DP-10))

  python test_recsys_assist.py --control_name Amazon_user_implicit_ae_0_genre_assist_constant-0.1_optim_0.5_dp-10

Results

• Results across all assistance rounds. Item-alignedDMTCDR outperforms 'Alone' baseline for both explicit andimplicit feedback.

Acknowledgements

Enmao Diao
Jie Ding
Vahid Tarokh