TPAB

NOTE We noticed that the first uploaded version had an error and did not apply popularity coarsening and bootstrapping. The July 26, 2025 version fixes this problem.

This repository contains the implementation of TPAB (Temporal Popularity Distribution Shift GeneralizABle Recommender System), as proposed in the paper "Generalizable Recommender System During Temporal Popularity Distribution Shifts" (KDD'25 V1): https://doi.org/10.1145/3690624.3709299.

---

Requirements

• Python 3.7
• PyTorch v1.13.0
• scikit-learn
• tqdm

---

Usage

Run the following command to execute TPAB and save the result file in the test_result folder:

python -u main.py --data_path 'micro_video' --dataset 'micro_video' --model 'mf' --algo 'tpab' --n_pop_group 20 --lambda1 1.0 --epochs 600 --decay 0.001 --lr 0.001 --recdim 64 --log_file './log/customized_file_name' --seed 2018 --gpu 0

Command Options:

• data_path/dataset: Specify the dataset. Supported options are:
  • 'micro_video'
  • 'kuairand'
  • 'yelp_10years'
• model: Backbone recommendation model:
  • 'mf'
  • 'lgn'
• algo: Method to use:
  • 'tpab' (Our method)
  • 'vanilla'
• n_pop_group: Coarsening parameter (K):
  • Examples: (K = -1, 10, 20, \dots)
  • (K = -1): No coarsening.
• lambda1: Bootstrapping parameter (\lambda):
  • Examples: (\lambda = 0, 1.0, 2.0, \dots)
  • (\lambda = 0): No bootstrapping.
  • Variants for ablation studies:
    • TPAB-T: algo='tpab-global', n_pop_group=X, lambda1=X
    • TPAB-C: algo='tpab', n_pop_group=-1, lambda1=X
    • TPAB-B: algo='tpab', n_pop_group=X, lambda1=0
• epochs: Maximum number of training epochs.
• decay: L2 regularization.
• lr: Learning rate.
• recdim: Embedding size.
• log_file: Log name (used for model files, logging, and result files).
  • This must start with './log/'.
• seed: Random seed for reproducibility.
• gpu: GPU ID to use for training.

---

Alternative Usage

You can also use the _tester.py script to run TPAB with custom hyperparameters:

python _tester.py

Modify the hyperparameters directly in _tester.py to configure the TPAB settings as needed.