Implementation for our paper Sinkhorn Transformer for Multi-Behaviour Recommendation.
The code is built on Pytorch and the RecBole library. Use following command to install the requeiremnts:
pip install -r requirements.txt
python runSTMBR.py --model=[STMBR] --dataset=[tmall_beh] --gpu_id=[0] --batch_size=[2048]
, where [value] is the default value.
- Note that we modified the evaluation sampling setting in
recbole/sampler/sampler.py
to make it static. - The model code is at
recbole/model/sequential_recommender/stmbr.py
. - Feel free to explore other RecBole library baseline models to compare the performances.