ToBE

This is the PyTorch implementation of Topology-guided Graph Transformer for Recommender Systems.

Requirements

• Environment:

  • Python 3.10.8
  • Nvidia RTX 4090 with CUDA 12.1

• Libraries:

  • torch == 2.1.2
  • torch_geometric == 2.4.0
  • numpy == 1.26.3
  • scikit-learn == 1.4.1.post1
  • scipy == 1.12.0
  • googledrivedownloader == 0.4

How to run

The commands and hyperparameter settings are provided in the run.sh script. The datasets will be automatically downloaded from the internet during program execution. Make sure the folder data exists in the root directory.

To reproduce the results reported in our paper, please follow the two steps below:

Step 1: Preprocessing

Run the following commands to preprocess the raw datasets. This step performs data filtering and precomputes the positional encodings for each node.

cd src/preprocess
for dataset in Yelp Gowalla AmazonBook Tmall Movielens1M AlibabaIFashion; do
    python3 preprocess.py --dataset $dataset --do_filtering
done

Step 2: Run Recommendation Model

After preprocessing, execute the recommendation model with the corresponding hyperparameter settings for each dataset:

cd src/task
python3 recommend.py --dataset Yelp --lr 0.0005 --lambda_reg 0.1 --lambda_au 0.5 --train_batch_size 2048 --num_layers 3
python3 recommend.py --dataset Gowalla --lr 0.001 --lambda_reg 0.1 --lambda_au 0.5 --train_batch_size 2048 --num_layers 3
python3 recommend.py --dataset AmazonBook --lr 0.001 --lambda_reg 0.1 --lambda_au 1 --train_batch_size 1024 --num_layers 3
python3 recommend.py --dataset Tmall --lr 0.001 --lambda_reg 0.1 --lambda_au 10 --train_batch_size 2048 --num_layers 3
python3 recommend.py --dataset Movielens1M --lr 0.0005 --lambda_reg 0.1 --lambda_au 0.5 --train_batch_size 2048 --num_layers 3
python3 recommend.py --dataset AlibabaIFashion --lr 0.001 --lambda_reg 0.2 --lambda_au 0.5 --train_batch_size 2048 --num_layers 3

Results