DeosciRec

Deoscillated Graph Collaborative Filtering. Paper in arxiv

Citation

@misc{liu2020deoscillated,
      title={Deoscillated Graph Collaborative Filtering}, 
      author={Zhiwei Liu and Lin Meng and Jiawei Zhang and Philip S. Yu},
      year={2020},
      eprint={2011.02100},
      archivePrefix={arXiv},
      primaryClass={cs.IR}
}

This code is based on Tensorflow 1.14 and python 3.6.

script to run DGCF on ml100k dataset:

python DGCF_osci.py --dataset ml100k --model_type DGCF_osci --alg_type dgcf --epoch 500 --regs [0.01]  --lr 0.001 --batch_size 512 --stop_step 5 --embed_size 64 --layer_size [64,64,64,64]

Notes

In order to train the model on large dataset, the arguement --low should be applied. For example, on ml1m dataset, one can use the script:

python DGCF_osci.py --dataset ratings_ml-1m --model_type DGCF_osci --alg_type dgcf --epoch 500 --regs [0.01]  --lr 0.001 --batch_size 1024 --low 0.01 --stop_step 5 --embed_size 64 --layer_size [64,64,64,64]

This argument is to filter the CHP laplacian matrix, where the value < low is filtered. More details can be find in the paper.

Acknowledgement

We reuse some part of the code in Neural Graph Collaborative Filtering https://github.com/xiangwang1223/neural_graph_collaborative_filtering